Mathematical Techniques in Structural Biology

MATHEMATICAL TECHNIQUES IN STRUCTURAL BIOLOGY
J. R. QUINE
Contents
0. Introduction 4
1. Molecular Genetics: DNA 6
1.1. Genetic code 6
1.2. The geometry of DNA 6
1.3. The double helix 6
1.4. Larger organization of DNA 7
1.5. DNA and proteins 7
1.6. Problems 7
2. Molecular Genetics: Proteins 10
2.1. Amino Acids 10
2.2. The genetic code 10
2.3. Amino acid template 11
2.4. Tetrahedral geometry 11
2.5. Amino acid structure 13
2.6. The peptide bond 13
2.7. Protein structure 14
2.8. Secondary structure 14
3. Frames and moving frames 19
3.1. Basic definitions 19
3.2. Frames and gram matrices 19
3.3. Frames and rotations 20
3.4. Frames fixed at a point 20
3.5. The Frenet Frame 20
3.6. The coiled-coil 22
3.7. The Frenet formula 22
3.8. Problems 24
4. Orthogonal transformations and Rotations 25
4.1. The rotation group 25
4.2. Complex form of a rotation 28
4.3. Eigenvalues of a rotation 28
4.4. Properties of rotations 29
4.5. Problems 30
5. Torsion angles and pdb files 33
5.1. Torsion Angles 33
5.2. The arg function 34
5.3. The torsion angle formula 34
5.4. Protein torsion angles. 35
5.5. Protein Data Bank files. 35
1
2 J. R. QUINE
5.6. Ramachandran diagram 36

5.7. Torsion angles on the diamond packing 37
5.8. Appendix, properties of cross product 38
5.9. Problems 38
6. X-ray Crystallography, Lattices, and Fourier Series 40
6.1. Lattices 40
6.2. The crystallographic restriction 41
6.3. Diffraction 42
6.4. Electron density 45
6.5. Fourier series 47
6.6. Three variable Fourier series 49
6.7. Diffraction pattern and Fourier series 50
6.8. Problems 51
7. Nuclear Magnetic Resonance and distance geometry 53
7.1. Larmor frequency 54
7.2. Splitting and chemical shift 55
7.3. Distance geometry 55
7.4. Obtaining coordinates from a distance matrix 58
7.5. Coordinates from the gram matrix 59
7.6. Problems 61
8. Discrete Frenet Frame 62
8.1. Discrete Frenet Frame 63
8.2. Definition of the discrete Frenet frame 63
8.3. Torsion angles and bond angles 64
8.4. The Euclidean Group 65
8.5. DFF and the Euclidean Group 66
8.6. Application to Proteins 66
8.7. Period 3 structures 69
8.8. Helix parameters 70
8.9. Problems 71
9. Protein Folding 73
9.1. Anfinsen’s hypothesis 74
9.2. Configuration space. 74
9.3. Reduced parametrizations. 75
9.4. Energy functions 76
9.5. Example, the alpha helix 79
10. Appendix 1: A Very Brief Linear Algebra Review 79
10.1. Matrices and Vectors 81
10.2. Operations with matrices 82
10.3. Matrix inverse 85
10.4. Dot product 85
10.5. Matrix multiplication and dot product 86
10.6. Symmetric Matrices 86
10.7. Determinant 87
10.8. Vector Cross Product 87
10.9. Eigenvalue and Eigenvectors 87
10.10. Rotation matrices 88
MATHEMATICAL TECHNIQUES IN STRUCTURAL BIOLOGY 3
10.11. Problems 89
4 J. R. QUINE
0. Introduction
These are notes about using mathematics to study the molecular structure of
molecules, especially long organic molecules like DNA and proteins. In both crys-
tallography and NMR structure determination, mathematics plays an important
role.
Distance geometry and discrete differential geometry are useful in structure de-
termination using NMR data. Molecules can be studied using matrices of distances
between pairs of atoms. The study of such matrices is called distance geometry.
Proteins and DNA are also long strings of atoms that can be modeled using differ-
ential geometry of curves. These techniques also apply to other questions in shape
theory and robotics.
Crystallography uses a different method. The molecule is modeled using level
surfaces of a positive function giving the electron density at a point in space. This
function is expanded as a Fourier series and the experiment gives information on
the coefficients of the series.
Section 1 discusses the structure of DNA. The discovery in 1953 of the double
helical structure of the DNA molecule by Watson and Crick initiated a revolution
in biology. First, it dramatically showed that the structure of a molecule can give
a clear picture of how it functions. In this case the double helical structure of DNA
shows how it replicates itself and transmits genetic information in the form of a
code. This helped initiate the subject of bioinformatics.
Secondly, the discovery by Watson and Crick of the structure of DNA showed
that molecular structures can be discovered by a considering chemical properties,
experimental data, and geometry. Ball and stick models and paper cutouts were
used to understand the geometry of the molecule. In the computer era, all the
relevant data is incorporated databases, and computers are able to manipulate
the structure. This is computational structural biology and molecular dynamics.
Section 9 gives a brief introduction to this.
There was a key piece of evidence that helped in the discovery of the structure
of DNA, an X-ray diffraction pattern found by Rosalind Franklin. The relationship
between the structure and the diffraction pattern was well known to Crick, and
he knew from the diffraction pattern that DNA was some kind of helix. Section 6
discusses how the electron density function of crystals is computed from diffraction
intensities using Fourier series.
Section 2 discusses proteins The discovery using X-ray crystallography of the
structures of the proteins myoglobin by John Kendrew in 1958 and hemoglobin
by Max Perutz in 1959 showed how the arrangement of the atoms in these struc-
tures reveals how these proteins function. There are a large number of atoms in
myoglobin, approximately 1200 not including hydrogens, and visualizing the ar-
rangement of them in 3D space is difficult. A graphic artist, Irving Geis, created
pictures to make the arrangement clear. The advent of computer graphics enabled
those studying proteins to see graphic renditions using ribbon diagrams and rotate
them with a mouse. The ribbon diagram is a graphic created by Jane S. Richard-
son to indicate a sequence of atoms in a helical formation. These helices are called
secondary structures. Secondary structures such as alpha helices, beta sheets, and
coiled coils are discussed in section 5 on torsion angles, in section 3 on the Frenet
formula, and in section 8 on the discrete Frenet frame.
Protein structures are stored in a database called the Research Collaboratory

for Structural Bioinformatics (RCSB) Protein Data Bank. Each protein whose
structure has been found has a file in which the coordinates in space of the center
of each atom is stored. This is called a pdb file and it is basically a list of atoms
and their coordinates in 3D space. A protein viewer program reads this file and
converts it to a graphic. At this writing there are about 86,000 files in the Protein
Data Bank. A description of these files is given in section 5.
In any discussion of shapes we are dealing with the geometry of space. The
most fundamental mathematical tool at our disposal is the group of Euclidean
motions. The sense preserving motions can be described in terms of rotations
and translations, and these can be put in matrix form. These transformations are
important in understanding symmetries such as the ones seen in crystals. The
simplest symmetries are given by translations and a structure with translational
symmetries is called a lattices. Lattices can also have rotational symmetries but
these are limited by the crystallographic restriction (section 6). Rotation matrices
can also be thought of as right-handed orthonormal frames. Frames and rotation
matrices are discussed in sections 3 and 4.
The fact that each peptide bond in a protein has the same geometry allows
us to consider the euclidean motion form one peptide bond to the next. These
trasformations are related to the Frenet formulas in the differential geometry of
curves.
In studying and manipulating protein structures, it is useful to have a parameter
space, a sequence of numbers which almost completely describes the structure. The
set of coordinates of the atoms is too large a set. Some distances and angles between
bonded atoms remain nearly the same for all proteins, and the backbone of a protein
is a long string of atoms. The result is that structure can be described by a sequence
of torsion angles. Torsion angles are discussed in section 5, and the way they are
used to describe a structure is discussed in section 8 on the discrete Frenet frame.
Two techniques for finding protein structures are X-ray crystallography and
NMR. Each technique has a set of mathematics tools used to find the structure
from the data. X-ray crystallography relies on the theory of space groups describe
the structure of the crystal, and on Fourier analysis to reveal the details of the
atoms in the unit cell. These ideas are described in section 6.
NMR uses distance constraints or orientational constraints. Distance constraints
are studied using distance and gram matrices described in section 7. Described there
is the technique of calculating a sequence of coordinates from a distance matrix.
This techniques is used in proving the Cayley Menger theorem and interpreting
Cayley Menger determinants. Orientational constraints are used in solid state NMR
and give information about the coordinates of the unit magnetic field directions in
the Frenet frames along the molecule. Discrete Frenet frames are discussed in
section 8. The Frenet formula derived in this section can be used to derive the
Denavit-Hartenberg formula used in robotics.
One of the major goals of structural bioinformatics is to find the structure of
a protein from the sequence of amino acids. Most techniques to accomplish this
require the use of an energy function on a configuration space for the protein and
a method of finding a configuration minimizing the energy. These ideas are briefly
discussed in the section on protein folding, section 9 and illustrate using Frenet
frames and the simple example of an alpha helix.
6 J. R. QUINE
1. Molecular Genetics: DNA

This section reviews a few basic facts about genetics and DNA. There is a lot
of information available from the Department of Energy. They have a primer on
molecular genetics available on the web. This is a short course in some techniques
in molecular biology used to find DNA sequences.
1.1. Genetic code. DNA is a long molecule consisting of a string of base-pairs.

Each base is a molecule called a nucleotide. There are four of these: Adenine,
Cytosine, Guanine, and Thymine. We abbreviate these with the letters A, C, G,
T. A long string of these four letters creates a code word, a gene, which gives
instructions for the manufacturing of a protein.
The reason that the code can be duplicated and passed along is that the nu-
cleotides pair only in certain ways: A pairs with T, and C pairs with G. These are
called complementary base pairs. When the string of base pairs is pulled apart,
we get two strings of complementary basepairs, each containing the same infor-
mation. The discovery that genetic information is coded this lead to a branch of
bioinformatics that investigates the similarity between long strings of letters A, T,
C, G.
The big breakthrough in thinking about genetic information as digital informa-
tion came with the landmark paper by Watson and Crick in 1953. This is a very
short paper and easy to read. It reminds you that some important discoveries are
based on simple observations. Knowledge of the geometry of molecules and chemical
bonds and some clever geometric thinking were used to find the structure of DNA
from only a few experimental facts. This is the prototype of how mathematics and
chemistry are used to discover facts about biology.
In this lecture we discuss the geometry of the DNA base pairing. Some of the
features of DNA structure are similar to protein structure, discussed in the next
chapter. For example, the formation of helices and hydrogen bonds is found in both
DNA and proteins. Also the pentagons and hexagons seen in the bases are similar
to the ones seen on protein side-chains.
1.2. The geometry of DNA. The reason that DNA functions the way it does is
due to the geometrical structure of the bases. In figure 1 are close up views of the
way nucleotides pair in a DNA structure.
The structure of DNA was discovered by Watson and Crick by studying the
geometry of the DNA basepairs. The shapes of the molecules and the distribution
of charges on them forces them to bond together only in pairs A-T and C-G. The
weak bonds between the bases, indicated by the dashed lines, are called hydrogen
bonds. The pair A-T is held together by 2 hydrogen bonds and the pair C-G by
3 hydrogen bonds. The bonds are caused by the electrical attraction between the
positive and negative charges on the atoms as indicated.
1.3. The double helix. The basepairs are strung together by a sugar-phosphate
backbone, and form a double helix. Figures 2 and 3 show in detail how the base
pairs fit into the helix.
A big clue for Watson and Crick was an x-ray diffraction photo (figure 4) of pu-
rified DNA fibers. This photo was obtained by Rosalind Franklin in the laboratory
of Maurice Wilkins. The inner cross pattern of spots indicated to Watson and Crick
that DNA was some sort of helix.
Figure 1. DNA basepairs showing the geometry of the hydrogen bonds
The relationship between the diffraction pattern and the structure of a molecule
will be discussed later in the book.
1.4. Larger organization of DNA. In the human organism, DNA is organized
into 23 pairs of chromosomes. Chromosomes are tightly wound strands of DNA.
The strands are very long and without the presence of certain enzymes, DNA
would become tangled and knotted. The way that DNA is ordered in chromosomes
and the way it is unknotted by the enzymes is a current topic of research which
uses ideas from geometry and topology.
1.5. DNA and proteins. The key to understanding the effect of genetics on health
and disease is to understand that genes code for proteins. It is these proteins which
determine the characteristics of organisms. Proteins and the genetic code will be
discussed in the next lecture.
1.6. Problems.
(1) From the Watson and Crick paper, what would be the approximate length
of a straight DNA helix of 106 basepairs?
8 J. R. QUINE
Figure 2. DNA structure

Figure 3. DNA helix indicating 10 basepairs per 360 degree turn

of the helix.
Figure 4. Rosalind Franklin’s x-ray diffraction picture of DNA

10 J. R. QUINE
2. Molecular Genetics: Proteins

2.1. Amino Acids. Proteins are long molecules composed of a string of amino
acids. There are 20 commonly seen amino acids. These are given in table 1 with
their full names and with one and three letter abbreviations for them. The capital
letters in the name give a hint on how to remember the one letter code
To a molecular biologist, each of these amino acids has its own personality in
terms of shape and chemical properties. Often the property can be given a numerical
value based on experimental measurements. One example, the hydropathy index,
is given in figure 5. This measures how much the amino acid dislikes dissolving in
water. An amino acid with a high hydropathy index, isoleucine, for example, can be
thought of as not mixing with water, or being oily. Some less precise classifications
simply divide the amino acids into two categories, hydrophilic (with low hydropathy
index) and hydrophobic (with high hydropathy index). Either of these classifications
can be used in computer programs for finding protein structures.
Alanine Ala A
Cysteine Cys C
Aspartic AciD Asp D
Glutamic Acid Glu E
Phenylalanine Phe F
Glycine Gly G
Histidine His H
Isoleucine Ile I
Lysine Lys K
Leucine Leu L
Methionine Met M
AsparagiNe Asn N
Proline Pro P
Glutamine Gln Q
ARginine Arg R
Serine Ser S
Threonine Thr T
Valine Val V
Tryptophan Trp W
TYrosine Tyr Y
Table 1. Amino acids and their abbreviations.
2.2. The genetic code. Based on the discovery of the structure of DNA as a long
word in a four letter alphabet, the key to genetics was found to be a code. A
sequence of three letters is a code for one of the 20 amino acids. A string of 3n
letters codes for a protein with n amino acids and gives the sequence in which the
amino acids are strung together.
Attempts were made to discover the code by logical reasoning, but the code was
found by experiments expressing proteins from manufactured sequences of DNA.
It was found, for example, that TGC codes for the amino acid Cystine. Also TAA
codes for STOP, which means that the string of amino acids stops, and the protein
is complete.
Figure 5. Figure from Introduction to Protein Architecture, by

Arthur Lesk showing a hydrophobicity scale for amino acids
2.3. Amino acid template. We are interested in the 3D structure of proteins.

Proteins are composed of amino acids bound together, so first we look at amino
acid structures. All amino acids have a COOH carboxyl and NHH amide part. The
part which distinguishes different amino acids is called the side chain or residue.
See figure 6
The structure of the amino acids can be learned by first learning their side chain
topology. The topology tells only how the atoms are connected; more information
is needed before you know the 3D structure. The additional information consists
of other parameters called torsion angles. Figure 7 (from a paper of Ponder and
Richards) gives the topology of the amino acids along with information on how the
atoms and torsion angles are labeled. The figure also indicates how many torsion
angles are needed to determine the structure. We will discuss this in more detail
in a later chapter and refer back to the figure often.
2.4. Tetrahedral geometry. The geometry of the amino acids is partially deter-
mined by the tetrahedral geometry of the carbon bond. The bond directions for
12 J. R. QUINE
Figure 6. Template for amino acid. R denotes the side chain,

or residue. The NHH is called an amide group and the COOH a
carboxyl group.
Figure 7. Side chain topology
carbon are approximately the same as from the centroid of a regular tetrahedron
to the vertices.
To get an idea of the geometry of a tetrahedron, a regular four-sided solid, you

can construct one in Maple. Here is a Maple file to construct a tetrahedron. The
tetrahedron can be rotated with the mouse.
The bond angles at a carbon bonded to four atoms are all approximately 110
degrees as if the carbon is the center of a tetrahedron and the bonded atoms are
at the vertices. To see this verify that the points A = (1, 1, 1), B = (1, −1, −1),
C = (−1, 1, −1) and D = (−1, −1, 1) are vertices √ of a tetrahedron by noting that
the
√ distance between any two of the points is 2 2. They are also all a distance of
3 from the origin so the origin is the center of the tetrahedron. Now the angle θ
between any two of the bonds, for example A and B, is given by
cos θ = A · B/|A||B| = −1/3.
Compute that θ = arccos(−1/3) ≈ 109.47◦ .
2.5. Amino acid structure. Figure 8 shows a typical structure of the amino acid
leucine. Configuration of side chains are sometimes called rotamers because the
Figure 8. Structure of leucine shown in a computer graphics stick

model. Hydrogens are white, carbons black, oxygens red, and ni-
trogens blue.
tetrahedral geometry at the carbon bonds stays the same and the only degree of
freedom is rotation about the carbon bonds.
2.6. The peptide bond. To form a protein, amino acids are bonded together in
sequence making a long chain. The bond between adjacent amino acids is called
the peptide bond. The carboxyl group of one amino acid and the amide group of
the subsequent amino acid lose an oxygen and two hydrogens, i. e., water (figure
9).
The bond is approximately planar; the six atoms involved in the bond lie in a
plane, called the peptide plane. The electrons associated with these atoms form a
cloud called the π orbital. There is a special geometry associated with the peptide
plane shown in figure 10.
14 J. R. QUINE
Figure 9. Peptide Bond. When two amino acids bond together

in forming a proteins, they give off one molecule of water.
Figure 10. Peptide plane geometry. a) shows the distribution of

electrons in the bond. b) shows the bond angles and distances as
determined from structures determined by X-ray crystallography.
This information is often used when modeling proteins.
2.7. Protein structure. As amino acids are bonded together they form into a
specific shape called the fold. The structure of a protein is hard to see because of
the number of atoms involved.
Before the era of computer graphics, only an artist could render an understand-
able picture of a protein. One such artist was Irving Geis. Here (figure 11) is his
painting of sperm whale myoglobin, the first protein structure to be discovered.
There is a website devoted to artistic renditions of molecules by Irving Geis and
others. These renditions have been replaced by computer graphics which rely on
ribbon diagrams.
2.8. Secondary structure. The organization of the atoms of a protein is complex

but certain regular features appear. The most common are the alpha-helix and the
beta-sheet (figures 12 and 13). These are referred to as secondary structures and can
be visualized using ribbon diagrams created by computer programs called protein
Figure 11. Painting by Irving Geis of a stick model of sperm

whale myoglobin.
viewers. These can be found online at the Protein Data Bank website. See figures
14 and 15. The spiraling ribbons are alpha-helixes and the straight ribbons are
beta-sheets. The original structure of sperm whale myoglobin by John Kendrew in
the the Protein Data Bank under the code 1mbn.
Figures 12 and 13 are a detailed view of the alpha helix and the beta sheet. The
structures are distinguished by the hydrogen bonding patterns. In an alpha helix
the hydrogen bonds join atoms nearby in the chain; in a beta sheet the hydrogen
bonds join atoms between two different parts of the chain.
16 J. R. QUINE
Figure 12. Stick diagram of the backbone of an alpha helix, show-

ing the hydrogen bonds in pink. Carbons are black, nitrogens blue,
oxygens red, and hydrogens white.
Figure 13. Diagram showing as dotted lines hydrogen bonds be-

tween protein strands in a beta sheet.
Figure 14. Ribbon diagram for the protein myoglobin. The

curled ribbons indicate alpha helices. There are no beta sheets
in this structure. The strings are called loops and have no partic-
ular structure.
18 J. R. QUINE
Figure 15. Part of the protein carboxopeptidase A containing

both an alpha helix and a beta sheet. The uncurled ribbons indi-
cate strands of beta sheets.
3. Frames and moving frames

In 3D space, a sequence of 3 linearly independent vectors v1 , v2 , v3 is called a
frame, since it gives a coordinate system (a frame of reference). Any vector v can
be written as a linear combination v = xv1 + yv2 + zv3 of vectors in the frame
and x, y, z are called the coordinates of v in this frame. A frame is the same as
an invertible linear transformation and an orthonormal frame is the same as an
orthogonal transformation.
A frame called the Frenet frame is useful in the study of curves. It is called a
moving frame because there is one at each point on the curve, and the points are
considered as a function of a parameter t often thought of as time. The idea of a
Frenet Frame can be adapted to study the shape of of long molecules such as DNA
and proteins, as will be discussed in a later chapter.
3.1. Basic definitions. A sequence of three vectors v1 , v2 , v3 can be made into

the columns of a 3×3 matrix denoted (v1 , v2 , v3 ). If the determinant of this matrix
is not 0, then the sequence of vectors is called a frame and the vectors are linearly
independent. We will not distinguish between the frame and the matrix.
The vectors
     
1 0 0
e1 =  0  e2 =  1  e3 =  0 
0 0 1
form a frame (e1 , e2 , e3 ) which is the identity matrix I. This is called the standard
basis or lab frame.
The determinant of the matrix (v1 , v2 , v3 ) is equal to the scalar triple product,
v1 · (v2 × v3 ) = det(v1 , v2 , v3 ).
If the determinant is positive, the frame is said to be right-handed, and if the

determinant is negative, the frame is said to be left-handed.
3.2. Frames and gram matrices. If F = (v1 , v2 , v3 ) is a matrix, the entry in

row i, column j of the matrix Ft F is the dot product vi · vj . This matrix is called
the gram matrix of the vectors v1 , v2 , v3 .
The frame is said to be an orthogonal frame if the vectors are mutually perpen-
dicular, that is, vi · vj = 0 for i 6= j. If all of the vectors are of length 1 in an
orthogonal frame, it is called an orthonormal frame. The condition that F is an
orthonormal frame can be written as
(
1 if i = j
vi · vj =
0 if i 6= j.
This is equivalent to the gram matrix being the identity,
(1) Ft F = I.
If F satisfies this condition it is said to be orthogonal. A frame that is not orthogonal

is called oblique.
20 J. R. QUINE
3.3. Frames and rotations. Vectors or points in space are usually given by col-
umn vectors whose entries are coordinates in the lab frame,
 
a
v =  b  = a e1 + b e2 + c e3 .
c
For simplicity a vector will often be written as a row vector (a, b, c) instead of the
column vector (a, b, c)t .
Any other frame F gives another set of coordinates to each point in R3 . If
F = (v1 , v2 , v3 ) then any w can be written as a linear combination of the vectors
in F as
(2) w = x v1 + y v2 + z v3
and (x, y, z) are said to be the coordinates in the frame F.
Equation (2) can also be written conveniently in matrix form as
 
x
(3) w = F y .
z
An orthonormal frame F can also be thought of as a way to specify an orthogonal
transformation. There is a unique transformation of space leaving the origin fixed
and sending the lab frame I = (e1 , e2 , e3 ) to an orthogonal frame F = (u1 , u2 , u3 )
and this is given by
t t
(x, y, z) → F(x, y, z)
so the transformation is given in lab frame coordinates by multiplication on the
t
left by the matrix F. If (x, y, z) are the coordinates in the lab frame before the
t
rotation, then F(x, y, z) are the coordinates in the lab frame after the rotation.
We see that
Fe1 = u1 Fe2 = u2 Fe3 = u3
are the columns of the orthogonal matrix F.
3.4. Frames fixed at a point. We have been thinking of frames as vectors with
initial points at the origin. If the initial points are fixed at a point p, the frame F is
denoted {F, p}. The Euclidean motion that sends {I, 0} to {F, p} is an orthogonal
transformation F followed by translation by p.
v → {Fv, p}.
3.5. The Frenet Frame. A moving frame is a frame F(t) which is a function of
t. The frame is thought of as changing with time and is often thought of as moving
along a curve. A curve in space is a vector function x(t) whose coordinates give
the parametric equations for the curve. Think of x(t) as the position of an object
at time t. If the initial points of the vectors of F(t) are at the point x(t) we write
{F(t), x(t)} and F is said to move along the curve x.
A moving frame frequently used with a curve is the Frenet frame. The first
vector in the frame is the unit tangent vector t pointing along the curve in the
direction of motion. The other vectors are the normal vector n and the binormal
vector b. The Frenet frame is defined as
F(t) = (t(t), n(t), b(t))
Figure 16. A moving frame (Frenet frame).
where
dx/dt
t(t) =
|dx/dt|
(4) dt/dt
n(t) =
|dt/dt|
b(t) = t(t) × n(t).
For the frame to be defined at a point we must have dx/dt 6= 0 and dt/dt 6= 0. The
speed is
ds/dt = |dx/dt| ,
so the unit tangent is the derivative of the position vector with respect to distance,
dx/dt
t(t) = = dx/ds.
ds/dt
By (4), (t, n, b) is a right handed frame, and the lengths of t and n are 1. To see
that it is an orthonormal frame, it is enough to show that t and n are perpendicular.
This is shown by differentiating
t·t=1
using the product rule, to get
2t · (dt/dt) = 0.
This is also geometrically evident, since t(t) can be thought of as a curve on a sphere
of radius 1. Any tangent vector to the curve at a point must be perpendicular the
radius of the sphere through the point, and therefore perpendicular to t.
3.5.1. Example of a Frenet Frame. Consider a helix of radius r and pitch 2πp where
the pitch of the helix is defined as the change in the z coordinate per turn of the
helix. The equation of the helix is
(5) h(r, p)(t) = (r cos t, r sin t, p t).
One complete turn of the helix corresponds to t going from 0 to 2π. During this
time the z coordinate changes by 2πp which is the pitch.
22 J. R. QUINE
The Frenet frame for the helix is computed by (4) to be

 
−c sin t − cos t s sin t
(6) F =  c cos t − sin t −s cos t ,
s 0 c
where
r p
c= p s= p .
r 2 + p2 r 2 + p2
Note that the normal n = (− cos t, − sin t, 0) is parallel to the xy plane and points
from the point h(t) towards the z axis which is the axis of the helix.
The Frenet frame of a helix given in (6) can be computed using Maple. Maple
demo
3.6. The coiled-coil. Francis Crick used a curve called the coiled-coil to describe
the backbone structures of proteins like collagen, which is found in fibrous tissue
in the body. He made his atomic structure by placing the alpha carbon atoms of a
protein at equally spaced parameter values along the curve. The equation for this
curve uses the Frenet frame for a helix.
The coiled coil is drawn by drawing a circle at a radians per second in the n, b
plane of the Frenet frame of helix (5) while traversing the helix. If this circle has
radius r2 then the vector equation of the coiled-coil is
C(t) = h(t) + r2 (cos at n + sin at b)
where h = h(r1 , p) is the equation of the helix. The equation can also be written
as
t
(7) C(t) = h(t) + r2 F(t)(0, cos at, sin at)
where F is the Frenet frame for h. The curve coils around a tube of radius r2 about
the helix h. In the moving frame the coil appears as a circle.
The helix h is called the major helix, and r1 is the radius of the major helix.
The coiled-coil C is called the minor helix, and r2 is the radius of the minor helix.
The major helix can be thought of as a curved axis for the minor helix. One turn
of the major helix corresponds to a variation of t by 2π; one turn of the minor helix
corresponds to a variation of t by 2π/a. Thus there are a turns of the minor helix
per turn of the major helix. (See figure 17 of a coil with with a = 4)
Maple demo of coiled coil
3.7. The Frenet formula. Curvature and torsion are defined by the derivative of
the Frenet frame. The derivative of a matrix is similar to the derivative of a vector,
it is computed by replacing each entry in the matrix by its derivative.
A square matrix A is said to be skew symmetric if
At = −A.
Note that that the entries on the diagonal of a skew-symmetric matrix are 0. An
important fact is
dA
If A(t) is orthogonal then At is skew-symmetric.
dt
This can be seen by taking the derivative of
At A = I
Figure 17. A coiled coil with 4 turns of the minor helix per
turn of the major helix.
using the product rule to get

dAt dA
A + At = 0.
dt dt
It follows that for the Frenet frame, Ft (dF/ds) is skew symmetric. In the case
of the Frenet Frame the 3,1 and 1,3 entries in the matrix are zero, because the 3,1
entry is b · (dt/ds) and dt/ds is parallel to n and therefore perpendicular to b. Now
curvature and torsion are defined by the Frenet formula,
 
0 −κ 0
dF
(8) Ft = κ 0 −τ 
ds
0 τ 0
where κ is curvature and τ is torsion. Since Ft F = I, equation (8) can also be
written
 
0 −κ 0
dF
(9) = F κ 0 −τ  .
ds
0 τ 0
Since
dF d t d n db
= , , ,
ds ds ds ds
equating the columns of (9) gives
dt
= κn
ds
dn
= −κt + τ b
ds
db
= −τ n.
ds
These are also called the Frenet formulas, but (8) is a convenient way to write them
in matrix form. It follows by taking the dot products of the first two equations with
n and b respectively that
dt dn
(10) κ= ·n τ= · b.
ds ds
24 J. R. QUINE
The curvature measures how the curve deviates from being a straight line. If
the curve is a line, then κ = 0. The torsion how it deviates from being in a plane.
If the curve is in a plane, then t and n are in the plane and b is the normal to
the plane, so τ = 0. Using the Frenet formula an equation for the curve can be
determined from the function κ and τ .
In the example of the helix (5) the curvature is a constant r/(r2 + p2 ) and the
torsion is constant p/(r2 + p2 ). In general the curvature and torsion are functions
of the parameter t.
3.8. Problems.
(1) Let
x(t) = (r cos t, r sin t, p t)
be a helix. Show that the curvature is a constant r/(r2 +p2 ) and the torsion
is constant p/(r2 + p2 ).
(2) Use Maple and the Frenet Formula
 
0 −κ 0
dF 
Ft = κ 0 −τ 
ds
0 τ 0
to find the curvature and torsion of the curve (3t − t3 , 3t2 , 3t + t3 ). You can
use the procedure presented in the lecture for computing the Frenet Frame
F.
(3) Show that
v1 · (v2 × v3 ) = det(v1 , v2 , v3 ).
4. Orthogonal transformations and Rotations

A matrix is defined to be orthogonal if the entries are real and
(11) A0 A = I.
Condition (11) says that the gram matrix of the sequence of vectors formed by
the columns of A is the identity, so the columns are an orthonormal frame. An
orthogonal matrix defines an orthogonal transformation v → Av by mutiplying
column vectors on the left.
Condition (11) also shows that A is a rigid motion preserving angles and dis-
tances. The dot product of two vectors is the same before and after an orthogonal
transformation. This can be written as
Av · Aw = v · w
for all vectors v and w. This is true by the definition (11) of orthogonal matrix
since
0
Av · Aw = (Av) Aw = v0 A0 Aw
= v0 Iw = v0 w = v · w.
Thus lengths and angles are preserved, since they are functions of the dot product.
The orthogonal transformation form a group since we can multiply two of them
and get an orthogonal transformation. This is because if A and B are orthogonal,
then A0 A = I and B0 B = I. So
0
(AB) AB = B0 A0 AB = I,
showing that AB is also orthogonal. Likewise the inverse of an orthogonal trans-
formation is orthogonal transformation.
Orthogonal transformations have determinant 1 or −1 since by (11) and prop-
erties of determinant,
(det A)2 = det(A0 ) det A
= det(A0 A)
= det I = 1.
4.1. The rotation group. Orthogonal transformations with determinant 1 are
called rotations, since they have a fixed axis. This is discussed in more detail
below. The rotations also form a group, since if det A = 1 and det B = 1 then
det(AB) = det A det B = 1.
If we think of an orthogonal matrix A as a frame
A = (v1 , v2 , v3 ),
then the determinant is the scalar triple product
v1 · (v2 × v3 ).
The frame is right handed if the triple product is 1 and left handed if it is -1. The
frame is the image of the right handed standard frame
(e1 , e2 , e3 ) = I
under the transformation A. Thus A preserves orientation (right-handedness) if
the determinant is 1.
26 J. R. QUINE
4.1.1. Rotations and cross products. Rotations are orthogonal transformations which
preserve orientation. This is equivalent to the fact that they preserve the vector
cross product:
(12) A(v × w) = Av × Aw,
for all rotations A and vectors v and w. Recall the right hand rule in the definition
of the cross product (fig 18). The cross product is defined in terms of lengths and
angles and right-handedness, all of which remain unchanged after rotation.
The fact that A preserves the cross product can also be proved using only the
fact that A0 A = I and det A = 1 (see problem 1).
Figure 18. The geometric definition of the cross product. The

direction of v×w is determined by the right hand rule. The fingers
of the right hand should point from v to w and the thumb in the
direction of the cross-product. The length of the cross product is
|v||w| sin θ .
4.1.2. Two dimensions. In two dimensions, every rotation is of the form

cos θ − sin θ
(13) R(θ) = .
sin θ cos θ
See section 10 problem 4.

Note that
R(θ)R(φ) = R(θ + φ) = R(φ)R(θ),
so that rotations in two dimension commute.
4.1.3. Three dimensions. In three dimensions, matrices for rotation about coordi-
nate axes have a form related to the 2 dimensional rotation matrices:
• Rotation about the x axis
 
1 0 0
Rx (θ) = 0 cos θ − sin θ
0 sin θ cos θ
Rotation about the y axis
 
cos θ 0 sin θ
Ry (θ) =  0 1 0 
− sin θ 0 cos θ
Rotation about the z axis
 
cos θ − sin θ 0
Rz (θ) =  sin θ cos θ 0 .
0 0 1
All rotations are counterclockwise about the axis indicated in the subscript. The
rotations Rz (θ), are exactly the ones that leave the vector e3 = (0, 0, 1)0 fixed and
can be identified with rotation in the xy plane. Similarly for Rx and Ry .
The rotations Rx (θ) commute,
Rx (θ)Rx (φ) = Rx (θ + φ)
= Rx (φ)Rx (θ).
Similarly rotations Ry (θ) commute and rotations Rz (θ) commute. In general, how-
ever, rotations in three dimensions do not commute. For example,
 
0 −1 0
Rx (π)Rz (π/2) =  −1 0 0 
0 0 −1
but  
0 1 0
Rz (π/2)Rx (π) =  1 0 0  .
0 0 −1
Since rotations form a group we can get other rotations by multiplying together
rotations of the form Rx , Ry , and Rz . It can be shown that any rotation A can be
written as a product of three rotations about the y and z axes,
(14) A = Rz (α)Ry (γ)Rz (δ).
The angles α, γ, δ are called Euler angles for the rotation A. See problem 6
The proof that any rotation A can be written as above in terms of Euler angles
relies on a simple fact. Any unit vector u can be written in terms of spherical
coordinates as
 
sin φ cos θ
(15) u =  sin φ sin θ  ,
cos φ
and u can be obtained from e3 by two rotations
(16) u = Rz (θ)Ry (φ)e3 .
28 J. R. QUINE
4.2. Complex form of a rotation. In dimension 2 it is convenient to use complex

numbers to write rotations. Rotation by an angle θ is given by
(17) ζ → eiθ ζ
where ζ = x + iy. Writing ζ̄ = x − iy the rotation

x cos θ − sin θ x
→
y sin θ cos θ y
is transformed into iθ
ζ e 0 ζ
→
ζ̄ 0 e−iθ ζ̄
which is convenient because the matrix is diagonal. The elements on the diagonal
are eigenvalues of
cos θ − sin θ
.
sin θ cos θ
A similar method works in three dimensions for rotation about the z axis. First
perform a change of coordinates. Letting ζ = x + iy the transformation
    
x cos θ − sin θ 0 x
 y  →  sin θ cos θ 0  y 
z 0 0 1 z
becomes    iθ  
ζ e 0 0 ζ
 ζ̄  →  0 e−iθ 0  ζ̄ 
z 0 0 1 z
transforming the matrix for the rotation into a diagonal matrix. The diagonal
entries are the eigenvalues of Rz (θ).
4.3. Eigenvalues of a rotation. First we show that the eigenvalues of an orthog-
onal matrix have absolute value 1.
To see this, suppose
(18) Av = λv
for a non-zero vector v. Taking the adjoint,
(19) v∗ A∗ = λv∗ .
Since A is real, A∗ = A0 , and multiplying (18) and (19) and using the fact that
A0 A = I get
v∗ v = |λ|2 v∗ v
and hence
|λ|2 = 1,
and thus all eigenvalues have absolute value 1.
If λ is an eigenvalue of A,
Av = λv
for some non-zero vector v. Taking conjugate of both sides,
Av̄ = λ̄v̄.
Since A is real, A = A so λ̄ is an eigenvector corresponding to eigenvector v̄. It
follows that the eigenvalues of an orthogonal matrix A are
±1 λ λ̄
where λ = eiθ . The determinant of A is the product of the eigenvalues, so for

a rotation matrix the first eigenvalue above is 1. There is a real eigenvector u
corresponding to the eigenvalue 1. This is left as an exercise (section 10 problem
6). The line though this vector u is called the axis of the rotation. .
By dividing by the length, we may suppose that u above is a unit vector. As in
(16) write u = Be3 where B is a rotation. Since u is left fixed by A,
ABe3 = Be3 .
−1
Thus B AB leaves e3 fixed and so
B−1 AB = Rz (θ)
for some angle θ. (See problem 2.)
We have shown that every rotation is conjugate to a rotation about the z axis.
In other words, every rotation A can be written in the form A = BRz (θ)B0 for a
rotation B and some angle θ. This is just a way of saying that by an orientation
preserving change of variables we can take the z axis e3 along the axis of the
rotation A. Since A and Rz (θ) are conjugate, they have the same eigenvalues, So
the eigenvalues of A are 1, eiθ , e−iθ , where θ is the angle of rotation. The vector u
is called the axis of the rotation.
Now we define the matrix R (u, θ), rotation an angle θ about the axis u. We
showed that every rotation A can be written
(20) A = BRz (θ)B0
where Be3 = u. In this case we write
A = R (u, θ) .
0
We can check that BRz (θ)B is the same for any choice of rotation B with Be3 = u
(problem 3), so R (u, θ) is uniquely defined. Note that R (e1 , θ) = Rx (θ), R (e2 , θ) =
Ry (θ), R (e3 , θ) = Rz (θ).
When (20) holds, A is said to be conjugate to Rz (θ). Equation (20) says that
under a change of coordinates given by a rotation B, R (u, θ) looks like Rz (θ).
Another way to think geometrically is that R (u, θ) is given by first rotating u to
the z axis, then rotating an angle θ about the z axis, then rotating the z axis back
to u.
4.4. Properties of rotations. A few of the main properties of rotations are sum-
marized here. In what follows, u is a unit vector and v and w are real vectors, and
A is a rotation,
(21) R (u, θ) = R (−u, −θ)
(22) Av · Aw = v · w
(23) Av × Aw = A(v × w)
(24) u · (R (u, θ) v − v) = 0
(25) AR (u, θ) A−1 = R (Au, θ)
(26) R (u, θ) v = cos θ (v − (u · v)u)

+ sin θ u × v + (u · v)u
30 J. R. QUINE
0
We can write (26) in a different form. If u = (a, b, c) write
 
0 −c b
Su =  c 0 −a
−b a 0
Then
(27) R (u, θ) = cos θ I + (1 − cos θ)uu0 + sin θ Su
The identities (22) was shown above. For (23), see problem 1. The proofs of
(24) and (25) are left as an exercise (see exercise 4).
Here is a proof of (26). The idea is to change variables so that u becomes e3 and
then the formula becomes obvious. Let A be a rotation such that Au = e3 and let
w = Av. Using (25) and applying A to both sides of the equation, it becomes
R (e3 , θ) w = cos θ (w − (e3 · w)e3 ) + sin θ (e3 × w) + (e3 · w)e3 .
Writing w = (x, y, z)0 the equation becomes
       
x x −y 0
R (e3 , θ)  y  = cos θ  y  + sin θ  x  +  0  ,
z 0 0 z
which is clear from the definition of R (e3 , θ).
4.5. Problems.
(1) (a) Show that if det A = 1, then
v1 · (v2 × v3 ) = Av1 · (Av2 × Av3 )
for all vectors v1 , v2 , v3 .
(b) If in addition A0 A = I, so that A preserves the dot product, show
that
A(v2 × v3 ) = Av2 × Av3
for all vectors v2 , v3 .
(2) Suppose A0 A = I and det A = 1 and Ae3 = e3 show that A is of the form
 
cos θ − sin θ 0
 sin θ cos θ 0 .
0 0 1
Use section 10 problem 13.
(3) Suppose B1 and B2 are rotations with B1 e3 = B2 e3 . Show that B1 Rz (θ)B−1
1 =
B2 Rz (θ)B−1
2 .
(4) Prove that if A is a rotation, then
AR (u, θ) A−1 = R (Au, θ) .
(5) Show that any rotation A can be written as a product of three rotations
about the y and z axes,
A = Rz (α)Ry (γ)Rz (δ).
The angles α, γ, δ are called Euler angles for the rotation A. (Hint: Write
Ae3 in spherical coordinates.)
(6) Every rotation can be written as a product of three rotations in the form
Rz (α)Ry (γ)Rz (δ), and the angles α, γ and δ are called Euler angles. Write
Rx (θ) in that form and find Euler angles for Rx . (The formula AR(u, θ) =
R(Au, θ)A might be √ useful.)
(7) Let u = (1, −1, 1)/ 3. Use (27) to find the matrix for R(u, 2π/3).
(8) If u = (a, b, c)0 write
 
0 −c b
Su =  c 0 −a .
−b a 0
Suppose that u is a unit vector (with real entries) so that |u|2 = u0 u = 1.
(a) Show that Su u = 0.
(b) Show that
Su2 = uu0 − I
(9) With Su as in the previous problem with u a unit vector,
(a) Show that
Su3 = −Su
(b) Show that
Su2n = (−1)n (I − uu0 ) for n ≥ 1
and
Su2n+1 = (−1)n Su for n ≥ 0
32 J. R. QUINE
5. Torsion angles and pdb files

In the study of space curves, the Frenet frame is used to define torsion and cur-
vature, and these are used to describe the shape of the curve. A long molecule such
as DNA or a protein can be thought of as a curve in space. Rather than being
described by continuous functions, it is described by line segments which represent
covalent bonds between atoms. The concept of curvature and torsion from differen-
tiable curves can be adapted to study the structure of these molecules. Curvature
corresponds to the angle between adjacent bonds, and torsion corresponds to the
torsion angle discussed here.
5.1. Torsion Angles. In the study of molecular structure, torsion angles are fre-
quently used to describe the shape of the molecule. In figure 19, we see four atoms
p1 , p2 , p3 , and p4 . Think of the vectors pj as vectors giving the coordinates of
Figure 19. Torsion angle φ = Tor (p1 , p2 , p3 , p4 ). The angle is

measured in the plane perpendicular to b = p3 − p2 .
the centers of the atoms. Let

(28) a = p2 − p1
b = p3 − p2
c = p4 − p3 .
and let Pa and Pc be the projections of a and c respectively onto the plane per-
pendicular to b. The angle, φ from −Pa to Pc, measured counterclockwise around
b, is the torsion angle. Denote this angle as
φ = Tor (p1 , p2 , p3 , p4 ) .
It is important to note that this angle is measured not between the two vectors −a
and c, but between their projections onto the plane perpendicular to b.
Since the torsion angle depends only on the vectors a, b, c also write
φ = τ (a, b, c) .
In this case the torsion angle is also called the dihedral angle. The angle is usually
measured in degrees and chosen in the interval (−180, 180].
The dihedral angle can be thought of as the angle between two planes (See
figure 20). It is the angle counterclockwise from the normal vector a × b of the
plane containing a and b to the normal vector b × c of the plane containing b and
c. Both a × b and b × c are in the plane perpendicular to b
MOLECULAR MODELING OF PROTEINS 411
p1
a
α p3
φ
p2 b β
c
p
FIG. 3. Bond vectors, bond angles, and the dihedral angle.
Figure 20. the The

specifying torsion
position angle
of the atom φ asIf two
in space. theatoms
angle
withbetween
labels j and planes.
k are
joined by a chemical bond, we consider the corresponding bond vector
φ = Tor (p1 ,xp
r = xk − j,
2 , p3 , p4 ) = τ (a, b, c) .
Angles αbond
with andlength
β are bond angles.
p
krk = (r, r) ,
where
5.2. The arg function. The torsion angle can be defined in terms of the argument
(p, q) := p1 q1 + p2 q2 + p3 q3
of a vector oriscomplex number. Define θ to be the argument of a vector (x, y) 6=
the standard inner product in R3 .
◦ i-j and k-l, we◦have the bond vectors
(0, 0), written θ = arg(x,
Similarly, for y), if −180
two adjacent bonds < θ ≤ 180 and
p = xj − xi ,pq = xl − xk .
cos θ = x/ (x2 + y 2 )
The bond angle α =<)(i-j-k), of Fig. 3, can pthen be computed from the formulas
sin θ
cos α =
= (p,y/
r) (x2 +kpy×2rk
, sin α =
).
kpkkrk kpkkrk
We can also write the argument in terms of complex numbers. The angle θ =
(together with α ∈ [0◦ , 180◦ ]), where
arg(x, y) if x + iy is written in polar form  
p2 r3 − p3 r2
r 
p × r =  p3 r1 − p1iθ
x + iyp1= rep2 r31.
r2 −
In Maple the iscommand

the cross product in R3 . The
to find thebond
argument of a complex
angle β =<)(j-k-l) number
is similarly found from is argument.
Note that if i, j, k is the standard (q,basis
r) then kqthe
× rk argument can be written as a
cos β = , sin β = .
kqkkrk kqkkrk
dihedral angle, ◦ ◦
Finally, the dihedral angle ω =<)(i-j-k-l) ∈ [−180 , 180 ] (or the complementary
(29) torsion angle 180◦ −arg(x, y) =
ω) measures the τrelative
(−i, orientation
k, xi + yj) . adjacent angles in
of two
a chain i-j-k-l of atoms. It is defined as the angle between the normals through the
The angle φ ofplanes
thedetermined
spherical by the atoms i, j, k andof
coordinates j, k,al, 3D
respectively,
vectorand can be calculated
from
(p × θ,
p ω==(cos × q) θ cos φ, sin
r, r sin (q ×θp,sin φ).
r)krk
cos , sin ω =
kp × rkkr × qk kp × rkkr × qk
can also be thought of the
In particular, as sign
a dihedral
of ω is given angle,
by that of the triple product (q × p, r).
(30) φ = τ (−i, k, p) .
5.3. The torsion angle formula. We give a formula for computing the dihedral
angle, hence the torsion angle, in terms of the argument.
The Dihedral Angle Formula. For vectors a, b, and c for which the torsion angle
is defined,
(31) τ (a, b, c) =
arg −|b|2 a · c + (a · b)(b · c), |b|a · (b × c) .

Proof. Recall that the left hand side is the angle from a × b to b × c measured
counterclockwise around b.
34 J. R. QUINE
Notice that both sides of (31) are unchanged if a, b, c are replaced by Aa, Ab,
and Ac for a rotation A. So we can assume b is in the direction of e3 . Likewise the
equation unchanged if b is replaced by λb for λ > 0 (dilation), so we can assume
b = e3 .
The equation is unchanged if a is replaced by its projection a − (a · b)b perpen-
dicular to b. So we can assume a is perpendicular to b. As above we can rotate
and dilate so that a = e2 . Let c = (x, y, z)0 . Then (31) is equivalent to
τ (e2 , e3 , (x, y, z)) = arg (−y, x)
which is true because the left hand side is the angle from e2 × e3 = e1 to
e3 × (x, y, z) = (−y, x, 0).
Here is a Maple worksheet to compute torsion angles.
5.4. Protein torsion angles.
5.4.1. Protein backbone torsion angles. The atoms along a protein backbone are
Cα -C-N-Cα -C-N-Cα . . . in a sequence repeating every third atom. If each atom has
a set of coordinates, the torsion angles along the backbone of a protein are named
as follows
• the angle Tor (C, N, Cα , C) is the φ torsion angle
• the angle Tor (N, Cα , C, N) is the ψ torsion angle
• the angle Tor (Cα , C, N, Cα ) is the ω torsion angle
Moving along the backbone we get a sequence of φ, ψ and ω torsion angles that
can be used to describe the structure of the backbone.
5.4.2. Protein sidechain torsion angles. We can also get torsion angles by moving
along a side chain. The greek letter subscripts for the atoms along the side chain
are indicated in figure 21. For example, the sequence of atoms Cα , Cβ , Cγ , Cδ of
Lysine determine the χ2 torsion angle. The atoms Cα , Cβ , Cγ , S of Methionine
determine the χ2 torsion angle.
For the χ1 angle, the first atom used for the torsion angle is the N on the
backbone. For example,
• for Leucine, the angle Tor (N, Cα , Cβ , Cγ ) is the χ1 torsion angle
• for Threonine, the angle Tor (N, Cα , Cβ , O) is the χ12 torsion angle (when
there are two χ1 angles, another subscript is added).
5.5. Protein Data Bank files. Structures of all known proteins are stored online
at the Protein Data Bank. The files there are called pdb files. The structural
information contained in the file is a list of three coordinates, (x, y, z), for the
centers of every atom in the molecule (although hydrogen atoms are left out because
they are small and their positions can be determined from the positions of the other
atoms). For example, a file identified as 1E0P contains the coordinates for a protein
called bacteriorhodopsin. Here is part of the file which can be downloaded from
the RCSB Protein Data Bank:
ATOM 1557 CB ILE A 205 -14.646 17.302 50.448 1.00 21.52 C
ATOM 1558 CG1 ILE A 205 -13.253 16.800 50.104 1.00 19.39 C
ATOM 1559 CG2 ILE A 205 -15.422 17.496 49.149 1.00 22.80 C
ATOM 1560 CD1 ILE A 205 -13.299 15.453 49.472 1.00 18.37 C
ATOM 1561 N PHE A 206 -12.336 19.006 52.127 1.00 22.92 N
Figure 21. Naming convention for torsion angles along the

sidechains of a protein
ATOM 1562 CA PHE A 206 -11.262 18.905 53.109 1.00 23.61 C

ATOM 1563 C PHE A 206 -11.438 19.982 54.171 1.00 26.34 C
The x, y and z coordinates are contained in columns 7, 8, and 9 respectively.
Other important information is
• column 2, the number of the atom in the list
• column 3, the position of the atom in the protein using the naming con-
vention in table (21). Note that, for example, CB is written instead of
Cβ .
• column 4, the three letter code for the amino acid
• column 5, the number of the amino acid in the list of amino acids (residues)
along the protein.
5.6. Ramachandran diagram. The φ, ψ pairs of torsion angles for each amino
acid along the backbone of a protein can be plotted as points in a rectangle, and this
plot is called a Ramachandran plot. The ω torsion angle in generally considered to
be 180◦ since the peptide bond is planar. The Ramachandran plot gives information
about the secondary structure of the protein. (See figure 5.6.)
A regular protein backbone structure is one where all of the ω torsion angles are
180 degrees and all of the φ, ψ pairs at alpha carbons have the same value. The
diagram below indicates the types of structures we get for different pairs. A regular
alpha helix, for example, corresponds to (φ, ψ) = (−60◦ , −50◦ ). The level curves
36 J. R. QUINE
Figure 22. The Ramachandran plot giving the number of

residues per turn of the helix.
indicate the number of residues per turn, the number of amino acids for each 180
degree turn about the axis of the helix, for the corresponding regular structure. All
of the structures can be thought of as helices with various numbers of residues per
turn.
5.7. Torsion angles on the diamond packing. The diamond packing is a set of
points in space where the centers of carbons of a diamond crystal lie. The diamond
packing is obtained from the face centered cubic lattice (the set of points with
integer coordinates adding up to an even number) by adding to it points of the face
centered cubic lattice moved over by the vector (1/2, 1/2, 1/2). So
Diamond packing = fcc ∪ {fcc + (1/2, 1/2, 1/2)} .
By moving on a path through the diamond packing you can get torsion angles of
180, 60, and −60 only (or undefined if two consecutive vectors are parallel). Since
the -60 degree torsion angles are close to the ones for alpha helices, attempts have
been made to model proteins by putting atoms in a protein on points in a diamond
packing. Maple demo
5.8. Appendix, properties of cross product. Here are some useful formulas
involving the cross product and the dot product
(a × b) · (c × d) = (a · c)(b · d) − (a · d)(b · c)
a × (b × c) = (a · c)b − (a · b)c
5.9. Problems.
(1) Suppose a pdb file contains the following lines
ATOM 813 N ARG O 113 -25.027 14.899 -17.838 1.00 8.68 N
ATOM 814 CA ARG O 113 -24.794 13.442 -17.751 1.00 8.96 C
ATOM 815 C ARG O 113 -24.526 13.069 -16.312 1.00 9.33 C
ATOM 816 O ARG O 113 -24.431 13.934 -15.454 1.00 8.24 O
ATOM 817 CB ARG O 113 -23.838 12.955 -18.888 1.00 7.62 C
ATOM 818 CG ARG O 113 -24.493 12.701 -20.214 1.00 7.02 C
ATOM 819 CD ARG O 113 -23.954 12.682 -21.559 1.00 4.59 C
ATOM 820 NE ARG O 113 -24.914 12.545 -22.637 1.00 4.39 N
ATOM 821 CZ ARG O 113 -24.964 12.763 -23.899 1.00 5.44 C
ATOM 822 NH1 ARG O 113 -24.081 13.156 -24.830 1.00 5.82 N
ATOM 823 NH2 ARG O 113 -26.153 12.663 -24.592 1.00 8.39 N
ATOM 824 N THR O 114 -24.536 11.769 -16.002 1.00 11.55 N
ATOM 825 CA THR O 114 -24.313 11.347 -14.633 1.00 11.92 C
ATOM 826 C THR O 114 -22.940 10.686 -14.493 1.00 11.89 C
ATOM 827 O THR O 114 -22.473 9.949 -15.295 1.00 11.88 O
ATOM 828 CB THR O 114 -25.467 10.527 -13.954 1.00 11.40 C
ATOM 829 OG1 THR O 114 -26.730 11.163 -14.288 1.00 10.92 O
ATOM 830 CG2 THR O 114 -25.425 10.484 -12.405 1.00 11.83 C
(a) Write in sequence the numbers of the atoms used to compute the χ2
torsion angle for ARG. Use Maple and the torsion angle formula to
compute the χ2 torsion angle.
(b) Using the coordinates above, use Maple to find (in degrees) the angle
Cα -C-O in THR.
(2) Using the formula in the notes for torsion angle, show that
Tor(p1 , p2 , p3 , p4 ) = Tor(p4 , p3 , p2 , p1 ).
38 J. R. QUINE
6. X-ray Crystallography, Lattices, and Fourier Series
Most of the information that we have on protein structure comes from x-ray
crystallography. The basic steps in finding a protein structure using this method
are:
• a high quality crystal is formed from a sample of protein

• the crystal is placed in an x-ray beam and the intensities of the diffraction
spots are measured
• after finding the phases, an electron density map is computed from the
diffraction intensities and phases using Fourier analysis. The phases cannot
be found from experimental data, but are guessed by methods not discussed
here.
• The coordinates of the atoms are found from the electron density.
• The structure is refined by checking that, for example, the atoms do not
get too close to each other.
What follows is a brief discussion of some of the mathematics involved in finding the
electron density from the diffraction intensities and phases. This requires studying
Fourier series for functions periodic on lattices.
6.1. Lattices. The basic structure of a crystal is that of a lattice. A crystal is

formed by several copies of the same protein in a pattern which fills a unit cell.
The unit cell is a parallelepiped which is then used as a tile whose translations fill
up space. For mathematical simplicity, we can suppose that there are an infinite
number of copies of the same protein.
A lattice is easy to describe mathematically. A lattice L in three dimensions

is generated by three linearly independent vectors a, b, c; it is the set of all points
h a+k b+lc where h, k and l are integers. A unit cell is the set of points xa+yb+zc
for 0 ≤ x ≤ 1, 0 ≤ y ≤ 1, 0 ≤ z ≤ 1. Similarly we can define a lattice in two
dimensions as the set of integer combinations of two linearly independent vectors
a and b and the unit cell is the parallelogram formed by the points xa + yb for
0 ≤ x ≤ 1, 0 ≤ y ≤ 1. In the pdb file for a crystal structure, you can find the
lengths of and angles between the vectors a, b and c which generate the lattice for
the crystal. If, for example, we look at the Protein Data Bank entry for the protein
2MBH, we find that the lengths of the vectors are a = 108.24. b = 63.13, c = 54.54.
The angles between the vectors are α = 90.00, β = 110.85, and γ = 90.00. The
angles are measured in degrees. The angle α is between b and c, the angle β is
between a and c and the angle γ is between a and b.
The origin of the coordinate system can be put at any point crystallized protein.
If the origin is placed at an atom in the protein, then every lattice point will be on
exactly the same atom in another copy of the protein in the lattice.
6.1.1. Examples of lattices. First consider two dimensions. The vectors a = (1, 0)
and b√ = (0, 1) generate the square lattice. The vectors a = (1, 0) and b =
(1/2, 3/2) generate the hexagonal lattice.
In three dimension the vectors a = (1, 0, 0), b = (0, 1, 0), and c = (0, 0, 1)
generate the cubic lattice. The vectors a = (1, 1, 0), b = (1, 0, 1), c = (0, 1, 1)
generate the face centered cubic lattice. The face centered cubic lattice can also be
described as the set of points (x, y, z) with integer coordinates such that x + y + z
is even.
For an illustration of lattices see the Maple demo.
The lattices described above are examples of lattices with symmetries. A sym-
metry of a lattice is a rotation such that the rotation and its inverse maps every
point in the lattice onto another point of the lattice. We say the rotation leave the
lattice unchanged. The square lattice, for example, is left unchanged by 90 degree
rotation, an order 4 symmetry since the product of 4 rotations of 90 degrees is the
identity. The hexagonal lattice is left unchanged by 60 degree rotation, an order 6
symmetry. The cubic lattice is left fixed by 90 degree rotation about any coordinate
axis. These are order 4 symmetries. The face centered cubic lattice is left fixed
by 120 degree rotation about the axis in the direction (1, 1, 1). This is an order 3
symmetry.
6.2. The crystallographic restriction. A symmetry of the lattice for a crystal

shows up in a symmetry of the diffraction pattern. It can be proved mathemati-
cally that the symmetry of a lattice has the crystallographic restriction: it has no
rotational symmetries of order 5 or greater than 6. The crystallographic restriction
can be stated as follows.
A lattice in the plane or in three-dimensional space cannot have rotational sym-

metries of order 5 or of order greater than 6.
Proof Here is a proof in dimension 2. The proof in dimension 3 is similar.
Let L be a two dimensional lattice generated by vectors a and b,
L = {h a + k b | h and k are integers} .

40 J. R. QUINE
Suppose that the rotation

cos θ − sin θ
(32) R=
sin θ cos θ
is a symmetry of L, that is, R moves L onto itself. Since R a and R b are in L,
R a = h1 a + k1 b
(33)
R b = h2 a + k2 b
for integers h1 , h2 , k1 , k2 .
Using the frame (a, b) write (33) as

R (a, b) = (a, b) M
or
(34) (a, b)−1 R (a, b) = M
where
h1 h2
M= .
k1 k2
The trace of a matrix A (tr A) is the sum of its diagonal entries. A basic fact in
linear algebra is that
trA = tr(B −1 AB)
for all invertible matrices B. Applying this to (34),
(35) tr R = 2 cos θ = tr M = h1 + k2 .
Now from (35) it follows that 2 cos θ is an integer, and so must be equal to −2,
−1, 0, 1, or 2. Thus the only possibilities for θ are 0◦ , ±60◦ , ±90◦ , ±120◦ , and
180◦ and this proves the crystallographic restriction in dimension 2.
6.3. Diffraction. Molecules arranged in a lattice are arranged in many sets of

parallel planes. You can see this by plotting a lattice, for example the face centered
cubic lattice, in Maple and rotating it (Maple demo).
An X-ray beam focused on a crystal creates a pattern of diffraction spots. The

position of an X-ray diffraction spot depends on the distance d between the parallel
planes creating it, the normal vector for the planes, and the wave length of the
x-rays.
The simplest and oldest rule of diffraction is Bragg’s law derived by the English
physicists Sir W.H. Bragg and his son Sir W.L. Bragg in 1913,
(36) 2 d sin θ = n λ
where θ is the angle of incidence of the x-ray, λ is the wave length of the x-ray, and
n is some integer. The formula holds when refraction yields a bright spot at angle
θ.
Figure 23. Illustration of Braggs Law. R1 and R2 are diffracted

X-rays. The angle θ is the angle of incidence with planes containing
the indicated atoms. Bragg’s law can be derived by looking at the
triangle ABC. The extra distance travelled by R2 is 2BC.
The formula is explained by looking at figure 23. If d is the distance between

the planes, the ray R2 travels a distance of 2 d sin θ farther than R1 . If this is an
integer multiple of the wavelength, the two rays are in phase after reflection and
create a bright spot. Otherwise there is cancellation and no spot is seen.
Here is a link to a Java applet illustrating Bragg’s Law.
6.3.1. Mathematical statement of Bragg’s law. Bragg’s law is an equation for the
phase difference between two sine waves. We say the functions sin 2π 2π
λ x and sin λ (x+
p
p) have phase difference λ cycles. The wave length is λ for both.
42 J. R. QUINE
A sine wave with amplitude 1 and wavelength λ moving in the direction of the
unit vector u1 can be written as

2π
(37) sin x · u1 .
λ
where x is a vector representing a point in space. This represents, at a fixed time,
an x-ray beam moving in the direction u1 . The reflection off a point x0 along a line
through x0 in the direction u2 , the wave has the equation

2π
(38) sin (x · u2 + x0 · (u1 − u2 )) .
λ
This can be seen by checking that (37) and (38) agree at x = x0 .
Similarly the reflection off the point x1 along a line through x1 in the same
direction u2 , the wave has the equation

2π
(39) sin (x · u2 + x1 · (u1 − u2 )) .
λ
See figure 24. The difference in phase between (38) and (39) is
Figure 24. x-rays of wavelength λ reflecting off two atoms at

points x0 and x1 . Incoming rays in direction u1 are parallel. Out-
going rays in direction u2 are possibly out of phase.
1
(40) (x1 − x0 ) · (u1 − u2 )
λ
cycles, and when the difference is a integer n,
(41) (x1 − x0 ) · (u1 − u2 ) = nλ,
the outgoing waves are in phase at the point at infinity in the direction u2 .
Suppose x0 and x1 are points in a lattice, so they differ by a lattice vector

x1 − x0 = v. Equation (41) shows that all such reflections are in phase if
(42) v · (u1 − u2 ) = nλ
for all lattice vectors v. For fixed n, (42) says that x0 and x1 lie in parallel planes
perpendicular to u1 − u2 .
Equation (42) is the mathematical expression of Bragg’s law (36). To see this,
note that (x1 − x0 ) · (u1 − u2 ) = d|u1 − u2 | and |u1 − u2 | = 2 sin θ.
It says, for example, that on the integer lattice, since v always has integer co-
ordinates, we get reflections in the direction u2 from the planes normal to u1 − u2
only if (u1 − u2 )/λ = (h, k, l) also has integer coordinates. The reflection spot in
the direction u2 is labelled (h, k, l) for integers h, k, and l. They come from planes
on the integer lattice parallel to the plane hx + ky + lz = 0.
6.4. Electron density. In crystallography, it is not the atoms of a molecule that

are seen, but the electrons surrounding the atoms. To describe an electron cloud,
a non-negative number giving the density of electrons is assigned to each point
in the crystal. This number can be thought of as the amount of charge per unit
volume. Although the information about protein structure is given in the form of
a file containing the coordinates of each atom, in reality what the crystallographer
sees is the electron density. The points representing the centers of the atoms are
extrapolated from the electron density function using a computer.
Suppose that a molecule is crystallized into a lattice consisting of all integer

combinations of vectors a, b and c. Let r(x, y, z) be the electron density at the
point x a + y b + zc, where x, y and z are real numbers, not necessarily integers.
The function r is called the electron density function. If a level surface of the
electron density function is plotted in one unit cell, it will look as in figure 25 for 6
Angstrom resolution and figure 26 for for 1A resolution.
The level surface shows more structural detail at higher (1 Angstrom) resolution.
The electron density map is the level surface of a function computed from a
Fourier series. Here is how Maple deals with level curves of functions of two variables
and level surfaces of functions of three variables.
The electron density function is given by a function of three variable which is

period 1 in each of the variables. Since the crystal is unchanged if it translated by
a, b or c, the electron density is the same at
(x + 1)a + yb + zc
as at
xa + yb + zc.
Thus r has period 1 in the variable x,
r(x, y, z) = r(x + 1, y, z).
A similar equation holds for the variables y and z. The electron density function is
a periodic function, of period 1 in each of the three variables.
44 J. R. QUINE
Figure 25. An electron density map of a molecule at 6 Angstrom

resolution. The atoms are not clearly seen.
Figure 26. Electron density function of a molecule at 1

Angstrom. The position of each atom is clearly seen.
Periodic functions are studied using Fourier series. We begin by looking at the
theory for one variable. This will generalize to the three variable case which we are
interested in.
6.5. Fourier series. The electron density function is constructed using Fourier
Analysis. Fourier analysis is the approximation of periodic functions by sines and
cosines. The basic idea of Fourier analysis is that any real valued function f (x) of
period 1 can be approximated by sums of the type
Xn
(43) a0 + [aj cos(2πjx) + bj sin(2πjx)]
j=1
for suitable choice of the real coefficients aj and bj and for n large enough. A sum
of type (43) is called a Fourier series.
The surprising part about Fourier’s discovery is that even a discontinuous func-
tion such as a square wave can be approximated by a Fourier series. For example,
consider the period 1 square wave defined on the interval [0, 1] by the function
0, if 14 < x < 43

(44) f (x) =
2, otherwise.
R1
This is a function of period 1, and 0 f (x) dx = 1. The Fourier cosine series
n
X
(45) 1+ aj cos 2π(2j + 1)x
j=0
with coefficients
4(−1)j
(46) aj =
π(2j + 1)
approximates f closely for large n. Here is a Maple demo
6.5.1. Complex form of Fourier series. We will discuss how the coefficients of a
Fourier series are computed. The simplest formula uses the complex form of the
Fourier series. Complex numbers give a convenient way of writing Fourier series,
even for real functions.
The complex form of Fourier series is based on Euler’s formula

eix = cos x + i sin x.
from which it follows that
e2πijx + e−2πijx
cos 2πjx =
2
and
e2πijx − e−2πijx
sin 2πjx = .
2i
Substituting these expressions in the general form for the Fourier sum (43) get
Xn
(47) cj e2πijx
j=−n
where
1
cj = (aj − ibj ) , j = 1, . . . , n
2
46 J. R. QUINE
c0 = a0
and
c−j = c¯j , j = 1, . . . , n.
So now we have a sum (47) of complex functions equal to the real sum (43).
The sum (47) adds up to a real function since every term cj e2πijx is added to its
conjugate cj e−2πijx .
If
n
X
(48) f (x) = cj e2πijx
j=−n
then the Fourier coefficients can be recovered from the function f by the formula
Z 1
(49) ck = f (x)e−2πikx dx.
0
This can be seen by multiplying both sides of (48) by e−2πikx and integrating,
noting that (
Z 1
2πi(j−k)x 6 k
0, if j =
e dx =
0 1, if j = k.
Here is a demo in Maple of complex Fourier series. It is left as an exercise

(problem 3 with a = 1/4) to show that formula (49) gives coefficients (46) for the
square wave (44).
The Fourier coefficients, like all complex numbers, have an absolute value and
an argument, that is, we can write them in polar form as
(50) cj = |cj |eiδj .
The number δj is called the phase and |cj | the norm. These are the phases referred
to in the crystallography phase problem.
6.5.2. Delta function. The square wave function (44) is an example of an approxi-
mation to a periodic 1 Dirac delta function. The periodic delta function δZ , or the
delta function of the one dimensional integer lattice Z, is thought of as a “function”
of period 1 such that
Z 1/2
(51) δZ (x)f (x) dx = f (0)
−1/2
for any smooth periodic function f (x). In particular,

Z 1/2
δZ (x) dx = 1.
−1/2
There is no function that has these properties, but the delta function can be thought
of as the limit as M → ∞ of square waves such as
M 1 1
(52) f (x) = 2 , if n − M < x < n + M , n an integer;
0, otherwise,
of which (44) is the special case M = 4. Similarly by replacing 0 by a in (51), get
the definition of the periodic delta function δZ+a .
A way to think about electron density function of a crystal and the phase problem
is to use a delta function. Think of the periodic delta function δZ+a in one dimension
as describing the electron density of a crystal formed by a single atom of zero
radius and mass 1 at each point a + n where n is an integer. Better, think of it
as approximated by a function which is non-negative but zero except near a and
which integrates near a to 1. By (49), the Fourier coefficients of δZ+a are given by
(53) cj = e−2πija .
So δZ+a can be approximated by
N
X
(54) δZ+a (x) ≈ e2πik(x−a)
k=−N
for large N . Look at the Maple worksheet to see what this function looks like. In
fact, simplifying the right hand side of (54) gives
sin 2π N + 12 (x − a)

(55) δZ+a (x) ≈ .
sin[π(x − a)]
The phase of cj is −2πja and the norm is 1. If we did not know the phase but
put 0 for the phase instead, then the coefficients cj would all be 1 and we would
get an atom at x = 0 instead of x = a. This illustrates the importance of knowing
the phases in computing the electron density function.
6.6. Three variable Fourier series. Fourier series can be used to analyze peri-
odic functions in any number of variables, for example the electron density function
ρ for a 3D crystal.
Given a real function ρ(x, y, z) of period 1 in each variable, it can be approxi-

mated by a triple sum of the form
Xn X n Xn
chkl e2πi(hx+ky+lz)
h=−n k=−n l=−n
written more compactly as

X
(56) f (x, y, z) = chkl e2πi(hx+ky+lz)
hkl
where
c−h ,−k ,−l = c̄hkl
48 J. R. QUINE
since the function is real, and where |h|, |j|, and |k| are ≤ n where n is a large
enough integer. Writing the complex Fourier coefficient in polar form as
chkl = |chkl |eiαhkl ,
the number αhkl is the phase.
Similar to (49) the Fourier coefficients of (56) are

Z 1Z 1Z 1
(57) chkl = f (x, y, z)e−2πi(hx+ky+lz) dx dy dz.
0 0 0
In x-ray crystallography, the values |chkl | are measured from the intensities of the
spots in the diffraction pattern. The phases αhkl must be supplied by other means
to find the electron density function. This is called the phase problem. There
are mathematical techniques for finding the phases but also some guess work is
involved.
6.6.1. Delta function. Similar to the delta function for the one dimensional integer
lattice, the delta function δZ3 for the three dimensional integer lattice, or cubic
lattice, can be approximated by a finite sum of the form (56) with chkl = 1 for all
integer triples hkl. Also δZ3 +a is approximated by setting the Fourier coefficients
c` = e−2πia·` where ` = (h, k, l).
6.7. Diffraction pattern and Fourier series. Here we show how the diffraction
intensities give the absolute values of the Fourier coefficients of the electron density
function. We show this when the molecule is a single point and the electron density
is a delta function. The general case follows from this by adding delta functions for
all the atoms in the molecule.
Consider the electron density of a single atom of mass m modeled approximately

as m times a delta function. For simplicity, assume the lattice of the crystal is the
integer lattice, Z3 . In order to get a finite Fourier sum consider just a part of the
lattice,
L = {(x, y, z)| x, y, z integers , −N ≤ x ≤ N, −N ≤ y ≤ N, −N ≤ z ≤ N }
for N large. Now write the electron density function for atoms at points x0 + `,
` ∈ L as
X
(58) ρ(x) = m e−2πix0 ·` e2πix·` .
`∈L
The Fourier coefficients of the electron density function are c` = me−2πix0 ·` for
` ∈ L. Note that ρ(x) is an approximation of the delta function δx0 +Z3 .
Now look at the diffraction pattern. As before, consider the reflection of an

x-ray beam moving in the direction u1 , reflected in the direction u2 from all of
these atoms. The equation for the beam reflected from the atom at x0 is given as
(38). The computation simplifies if we replace the sine function by the complex
exponential. Also assume that the intensity of the reflected beam is proportional
to the electron density m at the point. Then the outgoing beam has equation

2πi
(59) m exp (x · u2 + x0 · (u1 − u2 ))
λ
giving the amplitude and phase for the beam at a point x in space. The beam from
the entire lattice of points is the sum of the intensities of each point in the lattice,
(60)
X 2πi
w(x) = m exp (x · u2 + (x0 + `) · (u1 − u2 ))
λ
`∈L
X
2πi 2πi 2πi
= m exp x · u2 exp x0 · (u1 − u2 ) exp ` · (u1 − u2 ) .
λ λ λ
`∈L
Writing `∗ = (u1 − u2 )/λ equation (60) becomes

2πi
(61) w(x) ≈ m exp x · u2 exp [2πix0 · `∗ ] δZ3 (`∗ )
λ
since X
δZ3 (`∗ ) ≈ exp [2πi` · `∗ ] .
`∈L
Integrating it with respect to `∗ near a lattice point gives the value of 1. It follows
that the total reflected wave for (u1 − u2 )/λ near `∗ is

2πi
(62) m exp x · u2 exp [2πix0 · `∗ ]
λ
where x is a point along the reflected beam. (Since all the beams are parallel in the
direction u2 , we merge them all into one beam at infinity where they are collected.)
∗
The expression (62) is c`∗ exp 2πi λ x · u2 where c` is the ` Fourier coefficient of
∗
the electron density function (58). Since only the amplitude of the beam can be
measured from the intensity of the diffraction spot, the data gives only |c`∗ | = m,
the absolute value of the Fourier coefficient, and not the phase.
Using the above method of argument, it can be shown that the same result
holds when we add electron densities of atoms to get the electron density of a
molecule. Only the absolute value of the Fourier coefficient can be determined
from the diffraction pattern.
6.8. Problems.
(1) The face centered cubic (fcc) lattice is generated by the basis vectors
(0, 1, 1), (1, 1, 0), (1, 0, 1),
which means it is the set of all vectors of the form
a(0, 1, 1) + b(1, 1, 0) + c(1, 0, 1)
50 J. R. QUINE
where a, b, and c are integers. Show that this lattice is also the set of vectors
(p, q, r) where p, q, and r are integers and p + q + r is even.
(2) Show that the rotations
   
0 1 0 0 0 1
 0 0 −1  and  1 0 0 
−1 0 0 0 1 0
are symmetries of the fcc lattice.
(3) For 0 < a < 1/2 let fa (x) be the square wave defined on the interval
−1/2 ≤ x ≤ 1/2 by
(
1
if − a < x < a
fa (x) = 2a
0 otherwise
(a) Find the Fourier coefficients ck of the Fourier series
∞
X
ck e2πikx
k=−∞
for fa .
(b) Show that for every k, ck → 1 as a → 0. (Remark: the coefficients
ck = 1 are the coefficients of the delta function δZ )
(4) The Heaviside function can be used to construct square waves. Using Maple
plot the functions Heaviside(x) and
f (x) = Heaviside(1/4 − x) + Heaviside(1/4 + x) − 1
for −1 < x < 1.
(5) Use plot3d in Maple to plot the function
g(x, y) = f (x)f (y)
for −1 < x < 1, −1 < y < 1 and where f is the function in problem 4.
(6) For a function f (x, y) of period 1 in each variable the Fourier series is
defined by
N
X XN
ch,k e2πi(hx+ky)
h=−N k=−N
where Z 1 Z 1
ch,k = f (x, y)e−2πi(hx+ky) dxdy
0 0
are the Fourier coefficients. Let N be 2. Find the Fourier coefficients for
the function g in the previous problem. Use Maple to plot the Fourier series
and compare it with the graph of g. Try N = 3 and N = 4.
(7) Suppose
X n
f (x) = cj e2πijx
j=−n
with c−j = cj for all j so that f is a real valued function. Let

Z 1
g(x) = f (t)f (x + t) dt.
0
Show that
n
X
g(x) = |cj |2 e2πijx .
j=−n
Note: In crystallography the three dimensional analog of g is called the
Patterson function for f . It is determined by the norms of the Fourier
coefficients of f and not the phases.
7. Nuclear Magnetic Resonance and distance geometry
Nuclear Magnetic Resonance (NMR) is another method besides crystallography

that can be used to find structures of proteins. NMR spectroscopy is the observation
of spins of atoms and electrons in a molecule that is placed in a magnetic field. The
spins precess at a frequency in the radio frequency range and the frequency can be
detected by the electrical signal that it generates.
We briefly discuss the physics of NMR, and then describe distance geometry, a
mathematical theory that can be used to find the protein structure from some types
of NMR data.
7.1. Larmor frequency. Spins placed in a magnetic field precess; they wobble
like a spinning top. Only certain isotopes of molecules found in organic compounds
have spins that react to the magnetic field; the most common ones used in proteins
are 1 H, 13 C, 15 N. The isotopes 13 C and 15 N are not in common abundance, so
specially prepared protein samples must be used.
The frequency of precession is called the Larmor frequency and it is determined

mainly by the type of atom and the strength of the magnetic field. The basic NMR
equation is,
(63) ω = γB0
where
ω = the Larmor frequency, i.e., the angular
frequency of the precession in radians per second
γ = gyromagnetic ratio, a constant depending
on the type of the atom
B0 = the intensity of the magnetic field
7.2. Splitting and chemical shift. Equation (63) assumes that the detected fre-
quency depends only on the atom and the intensity of the magnetic field. There is
another factor, however. The magnetic field intensity B0 is not the same everywhere
in the molecule; it is affected by neighboring atoms and electrons. Neighboring spins
have their own magnetic field and this perturbs the field of the magnet and changes
the frequency of precession.
52 J. R. QUINE
Figure 27. An NMR experiment is the observation of the pre-

cession of nuclear spins in the presence of a magnetic field. The
large arrow represents the magnetic field of the magnet, the small
arrow represents magnetic field of the nucleus which is precessing
like a spinning top as the tip of the arrow moves on the indicated
circle.
This is illustrated by the NMR spectrum of the hydrogen atoms in the molecule
Toluene (figure 28). The spectrum is a Fourier transform of the electrical signal
showing the intensities (vertical axis) of certain frequencies (horizontal axis). The
spectrum can be thought of as the absolute values of Fourier coefficients for the
functionPgiving the signal as a function of time. It the signal is the real part of
n
s(t) = j=1 aj e2πiωj t then the spectrum gives absolute values |aj |, j = 1 . . . n at
frequencies ωj cycles per second.
The spectrum of Toluene shows that the peak frequencies cluster around two
values. Also a reference signal is shown for hydrogens which are not part of any
molecule. The frequencies of the hydrogens in the molecule are different (shifted)
from the reference signal. The observed difference is divided by the frequency of
the reference signal times 106 , and the change in frequency is reported in parts per
million or ppm. The change is called a chemical shift. The spectrum shows that
the 3 methyl hydrogens (to the right) are shifted less that the other 5 hydrogens.
From the symmetry of the molecule it is easy to see that the peak on the right
comes from the methyl hydrogens, the hydrogens on the CH3 group at the end of
the molecule. This is because they all have the same relation to the rest of the
molecule which causes the shift.
1
The 300 MHz H NMR spectrum of Toluene
methyl group
signal from
methyl hydrogens
signals from
non-methyl reference
hydrogens hydrogen
signal
d (ppm)
Figure 28. The spectrum of Toluene. In a magnetic field, the

hydrogens spins in the molecule precess at different frequencies
depending on the surrounding atoms and electrons. This is because
the neighboring atoms and electrons have spins which contribute
to the magnetic field. In Toluene the frequencies are seen in two
separate ranges shifted from the reference signal.
The spectrum of Toluene shows that we can infer facts about the shape of a
molecule by looking at the spectrum. This suggests that we can find chemical
structures of larger molecules by NMR. However the spectrum of a protein is much
more difficult to interpret. Below is the NMR spectrum of all the hydrogens in
the protein thioredoxin indicating which part of the molecule the hydrogen signals
are coming from. Although some aspects of the structure can be deduced from the
spectrum, it would be difficult to find coordinates of the atoms from this spectrum.
There is another type of spectrum (figure 30) called a 2D NOESY spectrum. This
experiment observes two frequencies from an atom, so the intensity is a function
of two variables. The figure shows level curves for high intensity, which looks like
a set of points. These can be used to estimate distances between atoms. Such
estimates are called distance constraints. Distance constraints can be used to find
atomic coordinates using techniques of distance geometry. Similar spectra can be
used to find orientational constraints which measure angles rather than distances.
Orientational constraint measure the angle between covalent bond vectors and the
vector giving the direction of the magnetic field.
54 J. R. QUINE
Figure 29. The hydrogen NMR spectrum of the protein thiore-

doxin. Indicated on the spectrum are the parts of the protein
responsible for each part of the signal. The signal is shifted ac-
cording to the unique chemical environment of each part of the
protein.
7.3. Distance geometry. Determination of the structure of a protein from dis-

tance constraints caused renewed interest in an old branch of mathematics called
distance geometry. A protein structure is a list of coordinates for the atoms. Rather
than coordinates, we could consider a list of distances between atoms. Distances,
unlike coordinates, are invariant under rotation and translation, so they are also
useful in shape analysis.
Consider a sequence of points in 3D space. Information about distances between

a sequence of points can be put into a matrix called a distance matrix. The atoms
in a structure are numbered 1 to n. The distance matrix is an n × n matrix. The
entry in row i and column j is the square of the distance between points i and
j. The object of distance geometry is to find coordinates of the points from the
distance matrix.
There is no unique list of coordinates since a rotated and translated set of coor-
dinates gives the same distance matrix, however, coordinates can be found which
F2(ppm)
Figure 30. NOESY spectrum of thioredoxin. A labeled point in

the spectrum indicates that the signal comes from the interaction
between a certain pair of hydrogen atoms. That the two hydrogen
atoms interact indicates that they are not far apart. The distance
between the atoms can be estimated by the intensity of the signal.
are unique up to a Euclidean motion. The situation is analogous to viewing a pdb

file. Protein viewers change all the coordinates in the structure through rotations
and translations, but the shape of the protein and the distances between atoms
stays fixed.
7.3.1. Distance matrix example. Label the rows and columns below by the letters
a through h. The corresponding entry gives the square of the distance between the
two corresponding points on the cube with sides of length 1 in figure 31.
 
0 1 2 1 1 2 3 2
1 0 1 2 2 1 2 3
 
2 1 0 1 3 2 1 2
 
1 2 1 0 2 3 2 1
 
1 2 3 2 0 1 2 1
 
2 1 2 3 1 0 1 2
 
3 2 1 2 2 1 0 1
2 3 2 1 1 2 1 0
For example, if va and vg are vectors giving the coordinates of points a and g
respectively, then the ag and ga entries in the matrix are |va − vg |2 . The distances
can be found from figure 31 or by writing out coordinates fro the vertices, va =
(0, 0, 0), vb = (0, 1, 0), etc.
56 J. R. QUINE
a b
e f
d c
g
h
Figure 31. Cube (octahedron) with vertices labelled.
7.4. Obtaining coordinates from a distance matrix. The problem in distance

geometry is to recover the coordinates of a sequence of points, up to a Euclidean
motion of space, from the distance matrix. The solution is presented in this section.
Suppose there are n + 1 vectors v0 . . . vn giving the coordinates of n + 1 points in

3D space. For simplicity, translate the points so that v0 = 0. Consider the vectors
as columns of a 3 × n matrix
M = (v1 , . . . , vn ).
Finding the vectors vj from the distance matrix follows in two steps
(1) find the gram matrix G = M0 M from the distance matrix. The matrix G
can be thought of as the square of M.
(2) find M from the gram matrix G. This can thought of as taking the square
root of the gram matrix.
Note that the distance matrix is (n + 1) × (n + 1) and the gram matrix is n × n.
7.4.1. Gram matrix from distance matrix. The entries in the distance matrix D are
|vj − vk |2 , j, k = 0, 1, . . . n,
|v1 |2 |v2 |2 |vn |2

 
0 ···
 |v1 |2 0 |v1 − v2 |2 · · · |v1 − vn |2 
 |v2 |2 |v2 − v1 |2 · · · |v2 − vn |2 
 
(64) D= 0 
 .. .. .. .. .. 
 . . . . . 
|vn |2 |vn − v1 |2 |vn − v2 |2 ··· 0
The gram matrix G is given by

 
v1 · v1 v1 · v2 ··· v1 · vn
 v2 · v1 v2 · v2 ··· v2 · vn 
(65) G=
 
.. .. .. .. 
 . . . . 
vn · v1 vn · v2 ··· vn · vn
The gram matrix can be found from the distance matrix using the identity
(66) − 2 vi · vj = |vi − vj |2 − |vi |2 − |vj |2 .
All the information on the right hand side of this equation is in the distance matrix
(64). Using (66) we can get the gram matrix from the distance matrix by row and
column operations
(1) subtract the first row of D from each row

(2) subtract the first column from each column.
The result is −2G bordered by zeros the first row and column.
7.5. Coordinates from the gram matrix. Given G obtained from a distance
matrix of vectors in 3D space, the problem is to find a square root of G, a 3 × n
matrix M such that
(67) G = M0 M.
This can be done by finding eigenvectors and eigenvalues. Since G can be written
in the form (67), it has real, non-negative eigenvalues. Since M is 3 × n there are
at most 3 non-zero eigenvalues. See exercise 1
Write the eigenvectors of G as columns of a matrix V and the eigenvalues as

diagonal entries in a diagonal matrix E. Then
(68) GV = VE.
Arrange the eigenvalues and eigenvectors so that

E1 0
(69) E=
0 0
where E1 is a diagonal matrix with 3 non-negative entries. Eigenvectors can always
be found which are an orthonormal set,
(70) V0 V = VV0 = I.
The eigenvectors are orthogonal if the eigenvalues are distinct and they can be
normalized by dividing by the length. If they are not distinct, an orthonormal set
of eigenvectors can be found by the Gram-Schmidt orthogonalization procedure.
Now (68) becomes
(71) G = VEV0 .
Write
(72) V = (V1 , V2 )
58 J. R. QUINE
where V1 is a 3 × n matrix whose columns are eigenvectors corresponding to the

diagonal elements of E1 , and where the columns of the 3 × (n − 3) matrix V2
have eigenvalue 0. The columns of V2 are in the kernel of G, that is, they are
eigenvectors corresponding to the eigenvalue 0, so GV2 = 0. Using (71) and (72)
we have
(73) G = V1 E1 V10 .
Now construct a diagonal matrix whose entries on the diagonal are the square
roots of the entries of E1 . Since the entries on the diagonal are non-negative,
√ the
square roots are also non-negative real numbers. Call this matrix E1 . From (73)
it follows that
p p
(74) G = V1 E1 E1 V10 .
Letting
p
M= E1 V10 ,
G = M0 M and M is a 3 × n matrix.
This procedure for finding coordinates is best illustrated using 4 points (n = 4).
Find coordinates of points in 3D space giving the distance matrix
 
0 2 1 1
 2 0 3 1 
 
 
 1 3 0 2 
 
1 1 2 0
Maple demo
Here is an example with n = 5. Find coordinates of points in 3D space giving

the distance matrix
0 2 1 1 2
 
 2 0 3 1 2 
 
 
 1 3 0 2 1 
 
 
 1 1 2 0 1 
2 2 1 1 0
Maple demo
Note that finding a sequence of vectors with the given distance matrix depends
on the gram matrix being rank 3 and having non-negative eigenvalues. If there are
small errors in the data, this will not be the case, but we can still solve the linear
algebra problem with vectors in a space of dimension n. These vectors may have
complex valued coordinates.
The above techniques are the basis for studying the Cayley-Menger theorem and
Cayley-Menger determinants.
7.6. Problems.
(1) Suppose G and M are real matrices with G = M 0 M .

(a) Show that all the eigenvalues of G are non-negative.
(b) If M is a 3 × n matrix show there can be at most 3 non-zero eigen-
values of G. (Hint: What is the dimension of the image of the linear
transformation M ? Same question for G.)
(2) Suppose M and N are 3 × 3 matrices whose columns are linearly indepen-
dent vectors. Suppose the gram matrices M 0 M and N 0 N are equal. Show
that there is an orthogonal matrix S such that M = SN .
(3) We say that A is a square root of G if A0 A = G. Using Maple and the
technique described in the notes, find a square root of
3.0 −1.0 0.0
 
 −1.0 3.0 −2.0  .

 
0.0 −2.0 6.0

To avoid complicated expressions, do this problem numerically.
(4) Using Maple, find coordinates of points in 3D space having the distance
matrix
0 4 4 8 3
 
 4 0 8 4 3 
 
 
 4 8 0 4 3 .
 
 
 8 4 4 0 3 
3 3 3 3 0
You can use commands from the worksheet DistMatrix4 on the website.
Give the answer to 2 decimal places.
8. Discrete Frenet Frame
We saw in the previous section how distance geometry is useful in finding protein
structures from NMR distance data. When the sample can be held rigid in relation
to the magnetic field, another method is used based on orientational constraints.
This method uses the NMR spectrum to find the coordinates of the unit magnetic
field direction in frames rigidly attached to the protein. These frames are discrete
versions of the Frenet frames discussed in section 3. We call such a frame a Discrete
Frenet Frame (DFF). The DFF is also useful in the study of robotics and kinematics
by transforming a rigid motions into a sequence of simpler ones.
8.1. Discrete Frenet Frame. In calculus, a curve in 3D space is given by a vector

function of a variable t. Using derivatives of this function the curvature and torsion
can be computed in terms of a Frenet frame, a moving frame along the curve, as
discussed in section 3.
60 J. R. QUINE
Organic chemistry includes the study of long molecules such as a proteins and
DNA. The backbone of a protein can be thought of as a sequence of points at the
center of atoms rather than as a continuous function of t. Using differences rather
than derivatives a Frenet frame can be defined which is useful in analyzing the
shape of the protein, and in finding protein structures using NMR orientational
constraints.
8.2. Definition of the discrete Frenet frame. A sequence of points v0 , . . . , vn

is called a discrete curve. To picture the curve, consecutive points are joined with
line segments, which in chemical applications are thought of as chemical bonds,
usually covalent bonds. The backbone of a protein, for example, can be thought of
as a discrete curve.
If no three consecutive points of a discrete curve are collinear, a sequence of

orthonormal, right-handed frames
Fk = (tk , nk , bk ) k = 1, . . . , n
where
vk+1 − vk
tk =
|vk+1 − vk |
tk−1 × tk
bk =
|tk−1 × tk |
nk = bk × tk
is defined called a discrete Frenet frame (DFF) for the curve. The unit vectors t,
n and b are analogous to the tangent, normal, and binormal for continuous curves.
The tangent vector tk points in the direction from vk to the next point vk+1 . The
binormal vector bk is perpendicular to the plane containing tk and tk−1 . Using the
cross product formula
a × (b × c) = (a · c)b − (a · b)c
the normal vector can be written as
−tk−1 + (tk−1 · tk ) tk
nk =
| − tk−1 + (tk−1 · tk ) tk |
so it is in the plane containing tk and tk−1 and perpendicular to tk .
8.3. Torsion angles and bond angles. Geometric properties of the curve can be
expressed in terms of the DFF. We have already defined the torsion angle for each
bond. The torsions angle at the bond from vk to vk+1 to be
φk = τ (tk−1 , tk , tk+1 ) k = 1, . . . , n − 1
The bond angle or curvature angle θk at vk is defined as the angle between the
bond vector from vk−1 to vk and the bond vector from vk to vk+1 ,
θk = arccos(tk−1 · tk ) k = 1, . . . , n
as in figure 32.
The bond angles and can also be defined using rotation matrices,
(75) tk = R (bk , θk ) tk−1
bk+1 = R (tk , φk ) bk .
Both of these equations follow directly from the definition of the bond angle and
the torsion angle.
q1
v
1
v
2
v
0
Figure 32. The angle θ1 is the bond angle at v1 . It is always

between 0 and 90 degrees.
Now write the rotation from Fk to Fk+1 in terms of the bond angle and the
torsion angle. By (75) and the fact that R (u, θ) leaves u fixed,
bk+1 = R (bk+1 , θk+1 ) R (tk , φk ) bk
(76)
tk+1 = R (bk+1 , θk+1 ) R (tk , φk ) tk .
Since rotations preserve cross products and nk+1 = tk+1 × bk+1 , it follows from
(76) that
nk+1 = R (bk+1 , θk+1 ) R (tk , φk ) nk .
The previous three equations can be combined as
(77) Fk+1 = R (bk+1 , θk+1 ) R (tk , φk ) Fk .
Now recall the property of rotations

(78) FR (u, θ) = R (Fu, θ) F.
where F is a rotation. Letting F = R (tk , φk ), u = bk , and θ = θk+1 get using (75)
and (78) that
R (tk , φk ) R (bk , θk+1 ) = R (bk+1 , θk+1 ) R (tk , φk ) .
Now (77) becomes
(79) Fk+1 = R (tk , φk ) R (bk , θk+1 ) Fk .
Since tk is the first column and bk the third column of Fk ,

Fk e1 = tk and Fk e3 = bk .
Now using (78) rewrite (79) as
(80) Fk+1 = Fk R (e1 , φk ) R (e3 , θk+1 )
= Fk Rx (φk ) Rz (θk+1 ).
62 J. R. QUINE
By induction using (80),

(81) Fk = F0 Rx (φ0 ) Rz (θ1 ) Rx (φ1 ) Rz (θ2 ) · · · Rx (φk−1 ) Rz (θk ).
So each frame can be written in terms of the the initial frame F0 and the sequence
of bond and torsion angles using the rotation matrices Rx and Rz .
8.4. The Euclidean Group. The orthogonal group is the group of all rigid trans-
formation of 3D space which leave the origin fixed. Every rigid transformation of
space T is given by an orthogonal transformation and a translation,
(82) Tw = Fw + v
where F is an orthogonal transformation and v is a vector. The set of all such
transformations is a group called the Euclidean group, denoted E(3), and an element
of this group is called a Euclidean transformation.
The composition of two Euclidean transformations

T1 w = F1 w + v1
T2 w = F2 w + v2
is given by
(83) T1 (T2 w) = F1 F2 w + (v1 + F1 v2 ).
The transformation (82) has inverse
(84) T−1 w = F−1 w − F−1 v.
A more convenient way to calculate composition of Euclidean transformations is

to use 4 × 4 matrices. Associate the transformation (82) with the 4 × 4 matrix

F v
(85)
0 1
where 0 denotes the zero row vector. Also associate the 3D vector w with the 4D
vector
w
.
1
Now
F v w Fw + v
=
0 1 1 1
and the vector on the right is associated with the vector (82). Likewise

F1 v 1 F2 v 2 F1 F2 v 1 + F1 v 2
(86) =
0 1 0 1 0 1
and the matrix on the right is associated with the transformation (83). So the com-
position of Euclidean transformations in 3D space can be computed by multiplying
the corresponding 4 × 4 matrices as in (86). For composition of Euclidean motions,
(86) can be easier to program than (83) since it relies on matrix multiplication.
The matrix (85) can also be thought as the fixed frame F whose vectors have
initial points at the point v.
8.5. DFF and the Euclidean Group. Equation (81) gives the frame Fk as a
function of the torsion angles and bond angles. The discrete curve cannot be
determined from the frames alone. Distances between points are needed to find
the points vk of the curve. Let sk = |vk+1 − vk | be the distance between two
consecutive points. Since tk = Fk e1 and
(87) vk+1 = vk + sk tk ,
the distances sk give the extra information needed to find the points, which can be
computed recursively from (87).
If the moving frame is thought of as an element of the Euclidean group in the

form (89), a formula can be written similar to (80),

Rx (φk ) sk e1 Rz (θk+1 ) 0
(88) Tk+1 = Tk .
0 1 0 1
where

Fk vk
(89) Tk =
0 1
represents the Frenet frame Fk at the point vk . This recursion formula gives fixed
Frenet frames at each point of the curve in terms of the initial frame, the bond
angles θk , torsion angles φk , and distances sk .
The bond angles, torsion angles and distances can be thought of as directions
for driving (or flying) along the curve. The distances sk between the points are like
the differences in odometer readings between pont k and k + 1. The bond angles
θk are like right and left turns if θk = π/2, except you may turn at any angle. Also
you do not need to make turns in the same plane, the torsion angle φk tells how to
move to a different plane.
8.6. Application to Proteins. Use of the DFF to study protein structure re-
quires knowledge of the geometry of the bonds in peptide plane. The atoms along
the protein backbone appear in sequence Cα , C, N, Cα , . . . and can be thought
of as points of a discrete curve. The bond angles at these atoms have been de-
termined by crystallography studies and the values are 58◦ at nitrogen atoms, 64◦
at carbonyl carbon atoms, and 69◦ at the alpha carbons (see figure 33). (If the
geometry at the alpha carbon is taken to be exactly tetrahedral, then the bond
angle arccos 31 = 70.5◦ is used.) Distances between atoms in the peptide bond have
also been determined. The distance from Cα to C is approximately 1.51, from C
to N approximately 1.32 and from N to Cα approximately 1.45.
8.6.1. Excluded regions in Ramachandran plot. In the Ramachandran plot, certain

φ, ψ pairs are excluded because they cause clashes between atoms in adjacent pep-
tide planes. Consider a pair of peptide planes bonded together with torsion angle
φ and ψ and with standard peptide geometry as indicated in figure 33. We com-
0
pute the distance d(O,O ) between the O atoms as a function of φ, ψ. The prime
indicates the second peptide plane. Think of the torsion angle pair (φ, ψ) as a con-
figuration space (or parameter space) for a dipeptide. The pair (φ, ψ) completely
64 J. R. QUINE
Figure 33. The geometry of the peptide plane as determined by

crystallography studies. All of the atoms indicated lie in the blue
plane.
k 1 2 3 4 5 6
vk O C N Cα C 0 O0
θk 56.5◦ 58◦ 70◦ 64◦ 0
◦
φk 0 0 φ ψ − 180 0
sk 1.24 1.325 1.455 1.51 1.24
Table 2. Bond angles, and torsion angles, and distance parame-
ters (in Angstroms) for the discrete curve from O to O0 in a dipep-
tide
determines the distance between all atoms. The protein segment from O to O0 is
a discrete curve consisting of atoms O, C, N, Cα , C0 , O0 . The bond and torsion
angles and distances along the curve are shown in table 2. Formula (80) can be
0
used find the coordinates of O and O0 and graph the distance d(O,O ) as a function
0
of φ, ψ (figure 34). The level curves of the function d(O,O ) shows regions of the
φ, ψ plane where the oxygens O and O0 are too close. Since the radius of an oxygen
atom is about 1.5 Angstroms, they certainly cannot get closer than a distance of
3. These excluded regions are shown on many Ramachandran diagrams (see figure
35).
Maple worksheet
8.7. Period 3 structures. The DFF can be used to study the helix parameters
for period three secondary structures, the beta sheet and the alpha helix. Suppose
0
Figure 34. Contour plot distances d(O,O ) between oxygens in
adjacent peptide planes as function of φ and ψ. Points where the
distance is less that 3 are excluded since the oxygens are too close.
Figure 35. Ramachandran plot showing all φ, ψ pairs (in yellow

boxes) from protein pdb ID 1AXC. Note from figure 34 the white
0
region with no φ, ψ pairs includes places where d(O,O ) is small.
66 J. R. QUINE
there are n peptide planes with atoms v−1 , v0 . . . v3n+1 with

v3k = the k th Cα atom
(90) v3k+1 = the k th carbonyl carbon atom
v3k+2 = the k th nitrogen atom.
The bond angles are given approximately by
(91) θ3k = 70◦ θ3k+1 = 65◦ θ3k+2 = 59◦
and the bond distances are
(92) s3k = 1.51 s3k+1 = 1.32 s3k+2 = 1.45
for k = 0, . . . , n. For a period three structure, the torsion angles also repeat in
sequences of three,
(93) φ3k = ψ φ3k+1 = 180◦ φ3k+2 = φ
where φ and ψ are fixed torsion angles. We discuss later the restrictions on φ and
ψ. The 180◦ angle is a consequence of the planarity of the three adjacent bones in
the peptide plane.
Initializing, let F0 = I and v0 = 0. From (81) it follows that

F1 = Rx (ψ)Rz (65◦ )
(94) F2 = F1 Rx (180◦ )Rz (59◦ )
F3 = F2 Rx (φ)Rz (70◦ )
Since the structure is period three, i.e., the angles repeat in units of three,
(95) F3k = Fk3 F3k+1 = Fk3 F1 F3k+2 = Fk3 F2 .
Letting Âk = {Fk ; vk }, (95) holds for F replaced by Â. So multiplication can
be done in the Euclidean group. The vector v3 is the vectors from alpha carbon 0
to alpha carbon 3 and is referred to as the virtual bond vector. It will be denoted
simply as v.
If T is any Euclidean transformation, and P is any point or set of points, we say

that the set of points T n (P ) is the orbit of P under powers of T . If P is a single
point, the orbit is called a helix since all of the points in the orbit will lie on a helix
curve whose axis is the axis of T . The above discussion shows that every peptide
plane in a period three structure is the orbit of the first plane under powers of Â3 .
The orbits of the atoms in the first peptide plane form helices. If the peptide plane
geometry is fixed, then Â3 is a function of φ and ψ only.
8.8. Helix parameters. To compute the parameters of protein secondary struc-

ture helices, the rotation F3 must first be written in the form R (u, θ). Remember
that u is a unit eigenvalue of the rotation F3 corresponding to the eigenvalue 1 and
θ = τ (w, u, F3 w) for any w not parallel to u. Also note that 1 + 2 cos θ is the trace
of F3 (sum of the diagonal entries).
There are two possible choices of u . The choice is made so that u · v > 0. The
helix is right or left handed depending on whether 0 < θ < 180◦ or −180◦ < θ < 0.
The helix parameters associated with the protein helix are

(96)
h = u · v = rise per residue
n = 2π/θ = number of residues per turn
p = hn = rise per turn

1 θ
r= cot u × v + v − (u · v)u = shortest vector from 0 to the helix axis
2 2
|vk − r − (u · vk )u| = distance of vk from the axis of the helix
If the alpha carbon is the origin of our coordinate system then
p
|v|2 − h2
(97) d = |r| =
2| sin θ2 |
= distance of alpha carbons from the helix axis.
Computing the products in (94), the helix parameters can be computed from
the torsion angle φ and ψ. It is convenient to let
φ+ψ φ−ψ
(98) s= t= .
2 2
Then
 
3.54
v = 1.37 cos ψ 
1.37 sin ψ
|v| = 3.80
θ
(99) cos = −.82 sin s + .03 sin t
2  
.82 cos s + .03 cos t
θ
sin u = −.57 cos s + .04 cos t
2
−.57 sin s − .04 sin t
θ
d sin = −.68 cos t + 2.9 cos s.
2
To derive the equations 99 it is convenient to use quaternions to represent the

rotations. For a unit traceless quaternion (vector) u define
eθu = cos θ + sin θ u.

The rotation from one alpha carbon frame to the next is given by
ψ α π β φ γ
(100) e 2 Ie 2 Ke 2 Ie 2 Ke 2 Ie 2 K
where in the standard peptide geometry, α = 59◦ , β = 65◦ , γ = 70◦ . We have
68 J. R. QUINE
(101)
α π β α β α β β−α α−β α−β
e 2 K e 2 I e 2 K = e 2 K Ie 2 K = Ie− 2 K e 2 K = Ie 2 K = cos I + sin J.
2 2
Since α and β differ by only 6 degrees, the above is close to I.
We compute (100) for the terms in (101) separately,

ψ φ γ γ
e 2 I Ie 2 I e 2 K = IesI e 2 K
γ γ γ γ
= I cos s cos + sin s cos I − sin s sin J + cos s sin K
2 2 2 2
γ γ γ γ
= − sin s cos + cos s cos I − cos s sin J − sin s sin K
2 2 2 2
and ψ φ γ γ
e 2 I Je 2 I e 2 K = JetI e 2 K
γ γ γ γ
= J cos t cos + sin t cos I − sin t sin J + cos t sin K
2 2 2 2
γ γ γ γ
= sin t sin + cos t sin I + cos t cos J − sin t cos K
2 2 2 2
so finally (100) is given by
γ γ
(102) cos δ I esI e 2 K + sin δ J etI e 2 K
α−β
where δ = 2 is near 3◦ . Now substitute
cos 3◦ cos 35 = .82
◦
cos 3◦ sin 35◦ = .57 sin 3◦ sin 35◦ = .03 sin 3◦ cos 35◦ = .04
to get the equations for cos θ2 and sin 2 u in (98). θ

8.9. Problems.
(1) Consider the discrete curve along the FCC lattice given by the 5 points
p1 = (0, 0, 0), p2 = (1, 1, 0), p3 = (1, 2, 1), p4 = (2, 2, 2), p5 = (3, 3, 2).
Compute the bond angles θ2 , θ3 , θ4 and the torsion angles φ2 , φ3 . (You can
write the torsion angle in terms of the arccosine function or the argument.)
(2) Consider the closed discrete curve given by the points
√ !0
jπ jπ j 5
vj = cos , sin , (−1)
3 3 4
for j = 0, . . . , 6. This models the discrete curve formed by the carbon
atoms in the chair configuration of the molecule cyclohexane. Note that
the molecule has symmetry. The set of points is left fixed by the rotation
Rz (2π/3).
(a) Find the discrete Frenet frame, F1 .
(b) Find the unit tangent vectors t0 , t1 , . . . , t6 .
(c) Show cos θj = − 31 , for j = 1, . . . , 6 so the bond angles are compatible
with a tetrahedral carbon bond.
(d) Find the torsion angles φ1 , . . . , φ6 .
Figure 36. Chair configuration of cyclohexane. Carbons are col-

ored black.
9. Protein Folding
The structure of proteins is described on several scales. The primary structure

is the amino acid sequence which is coded in the gene for the protein. Secondary
structure describes the chain forming alpha helices, beta sheets and loops. Tertiary
structure describes how the helices, sheets and loops are assembled into a working
unit. Review these concepts from the lecture on proteins.
Linus Pauling predicted the alpha helix secondary structure in 1948. His conjec-
tured structure was based on his knowledge of covalent bonds and hydrogen bonds,
and geometric reasoning using a piece of paper that he folded. Here is a copy of his
sketch on the unfolded sheet of paper. He asked himself how the long chain of the
protein backbone could be arranged so that the oxygens and the hydrogens on the
backbone form a hydrogen bond. His folded paper (figure 37) brought the points A
and B close together and created a hydrogen bond between the negatively charged
oxygen and the positively charged hydrogen. The hydrogen bond completed a ring
consisting of 13 atoms and showed a helix with 3.6 residues per turn. Structures of
proteins were later obtained by crystallography, confirming Pauling’s conjecture.
Figure 37. Paper folded by Linus Pauling in 1948 to predict the

structure of the alpha helix.
All Pauling needed to make his discovery was the sequence of atoms and a
knowledge about hydrogen bonds created by the electric force between the positively
70 J. R. QUINE
charged H bonded to N and the negatively charged. O bonded to C. This raises the
question of whether protein structures can be determined from just the sequence of
amino acids and all the forces between the atoms. Possibly with powerful computers
and a knowledge of all the forces between atoms in a protein, the structure can be
determined without experiments in the same way that Pauling did.
9.1. Anfinsen’s hypothesis. The above question is related to Anfinsen’s hypoth-

esis, one of the basic tenants of structural genomics. It states that information
determining the structure of a protein resides in the chemistry of its amino acid
sequence. For example, it says that the amino acid sequence of Bacteriorhodopsin:
TGRPEWIWLALGTALMGLGTLYFLVKGMGVSDPDAKKFYAITT
LVPAIAFTMYLSMLLGYGLTMVPFGGEQNPIYWARYADWLFTT
PLLLLDLALLVDADQGTILALVGADGIMIGTGLVGALTKVYSY
RFVWWAISTAAMLYILYVLFFGFTSKAESMRPEVASTFKVLRN
VTVVLWSAYPVVWLIGSEGAGIVPLNIETLLFMVLDVSAKVGF
GLILLRSRAIFGE
determines the structure shown in figure 38 that we find in the pdb file 1E0P.
Figure 38. The structure of the protein Bacteriorhodopsin
The structure of the functional protein is called the native state. The process by
which a protein folds into this structure from an extended chain is called folding
(figure 39).
9.2. Configuration space. The basic question in protein folding is whether from
the amino acid sequence and basic laws of physics the structure of a protein can be
predicted using the power of a computer. This leads to the following questions:
Figure 39. The relationship between protein sequence and pro-

tein function. Figure taken from Michael Levitt’s Structural Biol-
ogy and Computer Science course at Stanford.
• Is there a sequence of numbers that describes any possible shape, or con-

figuration, of a protein?
• What energy function can we associate to set of all configurations of a
protein with a given sequence of amino acids?
• Is the structure of a protein the solution to the problem of minimizing this
energy function on the set of configurations?
To describe the set of all possible shapes of a protein, we need a configuration

space described by a sequence of parameters. Recall that a structure is given
in a pdb file as a list of three coordinates for each atom in the molecule. This
configuration space is n dimensional where n is 3 times the number of atoms in
molecule. For bacteriorhodopsin we can see from the pdb file that one chain has
approximately has 1700 atoms. This gives n = 5100, a very large number of
variables to work with. The fact that proteins find their native configuration in
spite of the size of the configuration space is called Levinthal’s paradox.
Levinthal’s paradox: It would take a protein the present age of the universe to
explore all possible configurations and find the minimum energy configuration. Yet
proteins fold in microseconds.
9.3. Reduced parametrizations. For computation it is desirable to have a smaller

number of parameters describing configuration space. It is not necessary to find
72 J. R. QUINE
the coordinates of every atom. Some atom positions are fixed by others. For ex-
ample, if the coordinates of a carbon and two adjacent bonded atoms are known,
the coordinates of any hydrogen bonded to the carbon can be found.
The size of the configuration space can also be reduced by using torsion angles
for parameters instead of coordinates of atoms. The table of side chains is helpful
for counting. Below in table 3 is listed the number of torsion angles per side chain,
including main chain. On average there are about 4 (φ, ψ, χ) torsion angles per
residue.
Table 3. Number of torsions angle needed to determine the struc-
ture of a side chain.
number of
residues
torsion angles
2 Gly, Ala, Pro
3 Ser, Cys, Thr, Val
4 Ile, Leu, Asp, Asn, His, Phe, Try, Trp
5 Met, Glu, Gln
6 Lys, Arg
9.4. Energy functions. Pauling discovered the structure of the alpha helix by
trying to bring a hydrogen and an oxygen in close proximity while keeping the
geometry of the peptide bond and keeping some uniformity in the torsion angles
at the alpha carbons. If d is the distance from the O to the H in the hydrogen
bond, Pauling was trying to minimize d or equivalently to minimize −1/d. If q1
is the charge on the O, a negative charge, and q2 is the charge on H, a positive
charge, then by Coulomb’s law the electrostatic energy of the OH pair is q1 q2 /d. So
Pauling was minimizing energy. The numbers q1 and q2 are called partial charges
since they are less than 1 in absolute value, while the charge on one proton is +1
and the charge on one electron is −1.
This procedure can be greatly generalized by introducing the idea of an energy

function, a function of all the parameters for the configuration space, and trying
to find values of the parameters which minimize this function. One of the energy
functions used is the van der Waals energy which is large if two molecules are close
together. Minimizing the van der Waals energy keeps atoms from getting too close.
The van der Waals energy is just one term in the total energy.
Another term in the energy function is the electrostatic energy which is large if
two oppositely charged atoms are far apart. Minimizing the electrostatic energy
function will includes making the distance between atoms in a hydrogen bond small.
The electrostatic energy is another term in the total energy function. Many other
types of energy functions can be added to get the total energy.
One collection of energy functions that is frequently used is CHARMM (Chem-

istry at HARvard Molecular Mechanics). The functions are simple quadratic func-
tions of distances and angles, similar to the energy function for a spring. The
Figure 40. A spring in various states of displacement from equilibrium
energy can be thought of as the energy of a collection of springs describing the

forces within the molecule.
9.4.1. The energy of a spring. We review the equation for the energy of a spring.
If on a number line x = 0 represents the coordinate of the end of a spring at
equilibrium, the force on the spring with the end moved to x is given by F = −kx
where k > 0 is the spring constant(figure 40). The energy to move the end of the
spring to x is
Z x
x2
(103) E= ks ds = k .
0 2
Force represents the rate of change of energy, F = −dE/dx. We can think of the
force as a tendency to move to minimum energy (figure 41). The graph of the energy
function is called the energy landscape. To find the minimum energy configuration,
we look for valleys in the energy landscape. For this spring, the minimum energy
is at x = 0, at the lowest point of the parabola.
9.4.2. CHARMM Energy. The main part of the CHARMM energy is a sum of six
types of energy terms, each of which is a quadratic function analogous to the energy
function of a spring.
74 J. R. QUINE
Figure 41. Energy landscape of a spring. the curve represents

energy E = kx2 /2, with the minumum energy at x = 0. The force
is the slope of the energy function. The force pushes the spring
toward equilibrium at x = 0.
c` (b − b0 )2
P
V (x) = Pb b a bond length
+ Pθ ca (θ − θ0 )2 θ a bond angle
+ Pτ ci (τ − τ0 )2 τ a torsion angle
(104) + trig(ω) ω a dihedral angle
Pω Qi Qj
+ i,j D rij rij dist. between charged pair

P Ri +Rj
+ cw φ rij rij dist. between pair
Some further terms, accounting specifically for disulfide bonds and hydrogen
bonds, are also present but will not be discussed here.
9.4.3. Explanation of energy terms. The Qi are partial charges assigned to the
atoms in order to approximate the electrostatic potential of the electron cloud. The
constants labelled c are analogous to spring constants. These are estimated from
principles of physical chemistry and values for these are included in the CHARMM
package. D is the dielectric constant which is determined by the medium, usually
water, in which the protein folds. It measures how strongly within the medium an
electric charge is felt at a distance from the charge.
The quantities indexed by the subscript 0 are reference bond lengths, bond an-
gles, and improper torsion angles near their equilibrium values; different constants
apply depending on the chemistry and on their location in a functional group. The
coefficients of the trigonometric terms trig(ω) (linear combinations of cosines of
multiples of ω) are also determined by the chemistry of the atoms.
The van der Waals interactions (defined by the final sum in the potential) depend
on the interatomic pair potential ϕ which, in the simplest case, is taken as the
Lennard-Jones potential ϕ(R0 /R) where
ϕ(x) = x12 − 2 x6 .
Under this potential, two atoms are attracted if they are farther than R0 apart and
repel each other if they are less than R0 apart.
That all of these constants are independent of the molecule is a basic assumption
of molecular mechanics called transferability.
Slides from Michael Levitt’s structural biology course at Stanford are helpful in
understanding energy functions.
Finding the minimum of this complicated energy function is a hard computa-

tional problem.
9.5. Example, the alpha helix. As an illustration of minimizing an energy func-

tion, take a simple example of finding the φ, ψ parameters for an alpha helix as the
solution of a minimum energy problem. We saw that in the discovery of the alpha
helix, Pauling was trying to minimize d, or equivalently to minimize −1/d where d
is the distance between atoms forming a hydrogen bond. He was trying to find a
configuration giving the shortest distance between the H and the O atoms in the
bond. This can be thought of as minimizing the electrostatic energy given by the
charges on the O and H atoms.
Assuming that the O and the H have three Cα atoms in between, and assuming
that the φ, ψ torsion angles are the same at each Cα , the configuration space is 2
dimensional, consisting of the parameters φ and ψ. Looking at the discrete curve
from the O to the H atom, and using the discrete Frenet frame and the bond, torsion,
and distances given from protein geometry (see table 4), the distance d(O,H) can
be computed as a function of φ and ψ. Figure 42 shows the level curves of this
function. See the Maple worksheet for this computation.
10. Appendix 1: A Very Brief Linear Algebra Review
Linear Algebra, also known as matrix theory, is an important element of all

branches of mathematics. Very often in this course we study the shapes and the
symmetries of molecules. Motion of 3D space which leave molecules rigid can be
described by matrices. Briefly mentioned in these notes will be quantum mechanics,
where matrices and their eigenvalues have an essential role.
In all cases it is useful to allow the entries in the matrix to be complex numbers.
If you have studied matrices only with real number entries, it is very easy to adapt
to complex numbers. Almost all the rules of computations are the same.
76 J. R. QUINE
Table 4. The hydrogen bond completes a chain of 13 bonded

atoms along the protein forming a discrete curve. The bond angles,
torsion angles and distances along this chain from O to H are given
in this table.
k atom θk φk sk
1 O 1.24
2 C 56.5◦ 0 1.32
3 N 58◦ φ 1.45
4 Cα 69◦ ψ 1.51
5 C 54◦ 180◦ 1.32
6 N 58◦ φ 1.45
7 Cα 69◦ ψ 1.51
8 C 54◦ 180◦ 1.32
9 N 58◦ φ 1.45
10 Cα 69◦ ψ 1.51
11 C 54◦ 180◦ 1.32
12 N 60◦ 0 1.02
13 H
Figure 42. Level curves of d(O,H)(φ, ψ) showing a minimum near

(−50◦ , −50◦ ), which are the parameters for a right handed alpha
helix. There is also a minimum near (50◦ , 50◦ ) corresponding to a
left handed alpha helix.
In doing computations with matrices it is useful to have a computer program

such as Maple or Matlab. These tools make multiplication of matrices very easy,
and they work with complex numbers. The main difference between Maple and
Matlab is that Maple can work symbolically, that is, you can use letter as well as
numbers for entries. When using numbers, Matlab is often faster.
Below we give a review of a few basic ideas that will be used in the course.
10.1. Matrices and Vectors. Example:

2 1
A=
−1 3
A is a matrix with 2 rows and 2 columns i.e a 2 × 2 matrix.
A matrix with m rows and n columns is called an m × n matrix.
A matrix with the same number of rows and columns is called a square matrix.
3 × 3 square matrix:
 
3 1 7
B = −1 2 0
0 1 5
3 × 2 matrix:
 
2 0
C = −9 10
1 14
A 1 × 1 matrix is the same as a number or scalar,

3 = [3].
Matrices with 1 row are called row vectors and matrices with 1 column are called
column vectors.
 
3
2
A= B = 2
1
1
are column vectors.

C= 2 1 D= 3 2 1
are row vectors. Usually we will assume vectors are column vectors. A row vector
can be converted into a column vector (or vice versa) by the transpose operation,
which changes rows to columns.
78 J. R. QUINE
Example:
 
0 1
1 −1 0 = −1
0
Complex numbers can be used in matrices.
A number z = a + bi where a and b are real numbers and where i2 = −1 is

called a complex number. The number a is called the real part, a = Re z and b the
imaginary part, b = Im z.
If z = a + bi where a and b are real numbers, then the complex conjugate of z is

z = a − bi. The number z is real if b = 0, or equivalently, z = z̄.
Basic operations with matrices are:
• addition
• scalar multiplication
• multiplication
10.2. Operations with matrices. Addition: Add matrices by adding correspond-

ing entries.

2 1 i 0 2+i 1
+ =
−1 3 1 2 0 5
Every matrix entry is multiplied by the scalar.

2 1 4i 2i
2i =
−1 3 −2i 6i
Matrix Multiplication: Multiplication AB can be done only if the number of

columns of A is the same as the number of rows of B. Each entry of the product is
the dot product of a row of the first matrix with a column of the second.
Here is the product of a 3 × 3 matrix and a 3 × 1 matrix.

      
2 1 3 1 2·1+1·0+3·2 8
−1 1 2 0 = (−1) · 1 + 1 · 0 + 2 · 2 = 3
3 1 1 2 3·1+1·0+1·2 5
Here is the product of two 3 × 3 matrices:

  
2 1 3 1 −1 0
1 0 1 1 2 1
2 =
2 1 0 3 0 1
1
 
2 · 1 + 1 · 1 + 3 · 3 2 · (−1) + 1 · 2 + 3 · 0 2 · 0 + 1 · 2 +3·1
1 · 1 + 0 · 1 + 1 · 3 1 · (−1) + 0 · 2 + 1 · 0 1 · 0 + 0 · 1
2 + 1 · 1 =
1
2 · 1 + 1 · 1 + 0 · 3 2 · (−1) + 1 · 2 + 0 · 0 2 · 0 + 1 · 2 +0·1
 
12 0 3.5
 4 −1 1 
3 0 .5
An important thing to remember about matrix multiplication is that it is not

commutative; in general, AB 6= BA. For example

1 1 1 0 2 1
=
0 1 1 1 1 1

1 0 1 1 1 1
=
1 1 0 1 1 2
Other operations are
• conjugation
• transpose
• adjoint
• inversion
If Z is a matrix then the matrix conjugate is formed by taking the complex

conjugate of each entry.
Example: Let

1+i 2 1 2 1 0
Z= 1 = 1 + i
3−i 2 + 2i 3 2 −1 2
= A + Bi
where
1 2
A= 1
3 2
and
1 0
B=
−1 2
.
The matrix A is the real part of Z and the matrix B is the imaginary part of Z.
80 J. R. QUINE
The conjugate of Z is
1−i 2
Z=
3 + i 12 − 2i
or using the real and imaginary parts of the matrix,
Z = A − Bi
Some properties of the matrix congugate are:
AB = A B
(A + B) = A + B
The transpose of a matrix A, A0 , is obtained by changing rows to columns (or

equivalently, changing columns to rows).
Sometimes the transpose is denoted A0 rather than At .
   
2 1 −1 2 0 −1
A= 0 1 2 At =  1 1 0
−1 0 1 −1 2 1
Some properties of transpose are
t
(AB) = Bt At
t
(A + B) = At + Bt
The Hermitian transpose or adjoint is the conjugate transpose given by

t
A∗ = Ā .
Example:  
∗ 1
1, 2 − i, 3 = 2 + i
3
Example: For the matrix Z given above,

∗ t 1−i 3+i
Z =Z = 1
2 2 − 2i
Some properties of adjoint are

(AB)∗ = B∗ A∗
(A + B)∗ = A∗ + B∗
10.3. Matrix inverse. The matrix I denotes a square matrix whose entries are aij
where (
1 if i = j
aij =
0 if i 6= j.
The matrix I is called the unit or identity matrix. Identity matrices come in different
sizes:  
1 0 0
1 0
I2 = I3 = 0 1 0 .
0 1
0 0 1
Let A be a square matrix. The inverse is a matrix A−1 such that A−1 A =
AA−1 = I.
The inverse of a 2 × 2 matrix is easy to find. If

a a12
A = 11
a21 a22
then
1 a22 −a12
A−1 =
α −a21 a11
if α = a11 a22 − a12 a21 6= 0.
Not every square matrix has an inverse.
   
a x
10.4. Dot product. Let v =  b  and w = y  be two vectors, then the dot
c z
product is given by
 
x
v · w = v t w = a b c y  = ax + by + cz

z
   
1+i 2
Let v =  1  and w = 1 − i then the Hermitian dot product is given by
i 3
hv, wi = v ∗ w = v · w
= (1 − i)2+1(1 − i) + (−i)3
= 3 − 6 i.
82 J. R. QUINE
The length of a vector |v| is given by

|v|2 = hv, vi.
Two vectors v and w are orthogonal or perpendicular if hv, wi = 0. In general

for real vectors
hv, wi = |v||w| cos θ
where θ is the angle between the vectors.
10.5. Matrix multiplication and dot product. An m × n matrix can be con-

sidered as a list of n column vectors, v1 , . . . , vn , each m × 1,
A = [v1 , . . . , vn ]
and the transpose as a list of row vectors
v1t
 
At =  ...  .
 
vnt
If
w1t
 
B =  ... 
 
wkt
is a k × m matrix given as a list of k row vectors, w1t , . . . , wkt , each 1 × m, then
 
w1 · v1 . . . w1 · vn
BA =  ... .. 

. 
wk · v1 ... wk · v n
is a matrix of dot products.
10.6. Symmetric Matrices. A matrix B is symmetric if B = Bt
Example:  
3 1+i 2
1 + i 0 −5
2 −5 2
is symmetric.
A matrix B is self adjoint (or Hermitian symmetric) if B∗ = B
Example:
1
 
2 1+i 2 + 2i
 1−i 3 5 
1
2 − 2i 5 4
is self adjoint.
10.7. Determinant. The determinant of a 2 × 2 matrix is given by

a b a b
= ad − bc.
det =
c d c d
For a 3 × 3 matrix,
 
a b c
e f

d

f d e
det d e f = a
 − b + c .
h k g k g h
g h k
This is called expansion by minors. Likewise the determinant is defined for any
square matrix.
Properties of determinant:
detAB = detA detB
det A 6= 0 implies A−1 exists
10.8. Vector Cross Product.

     
a x bz − cy
 b  × y  = −az + cx .
c z ay − bx
= (bz − cy)i − (az − cx)j + (ay − bx)k

where      
1 0 0
i = 0 j = 1 k = 0
0 0 1
The formula for cross product is often remembered by pretending that i, j and
k are numbers and writing
     
a x i j k
 b  × y  = det a b c 
c z x y z
10.9. Eigenvalue and Eigenvectors. The scalar λ, a real or complex number, is

an eigenvalue of a matrix A corresponding to an eigenvector v 6= 0 if Av = λv.
Example: 1

2 1 1 1
=3
1 2 1 1
The eigenvalue is 3, and an eigenvector is [1, 1]t . Note that [2, 2]t is also an
eigenvector.
84 J. R. QUINE
Example: 2

0 −1 1 i 1
= =i
1 0 −i 1 −i
The eigenvalue is i, and an eigenvector is [1, −i]t .
Example: 3

0 −1 1 −i 1
= = −i
1 0 i 1 i
The eigenvalue is −i , and an eigenvector is [1, i]t .
Note that the equation in example 3 is the conjugate on the one in example 2.
Also note that a matrix with real entries can have complex eigenvalues and
eigenvectors.
The eigenvalues of self adjoint matrices are real. This fact is essential in many
areas of mathematics and is also a key fact in the mathematical formulation of
quantum mechanics. Here is a proof:
If A is self adjoint and

Av = λv
then taking the adjoint of both sides gives since A∗ = A,
v ∗ A = λv ∗ .
Multiplying the first equation on the left by v ∗ and the second on the right by v
gives
v ∗ Av = λ|v|2 = λ|v|2 .
Since |v| =
6 0 we have λ = λ and so λ is real.
A consequence is that the eigenvalues of a symmetric real matrix are real.
10.10. Rotation matrices. In dimension 2, rotation of a column vector by an

angle θ counterclockwise is given by multiplying on the left by the matrix

cos θ − sin θ
.
sin θ cos θ
In dimension 3 rotations about the three axes are given by

• Rotation an angle θ about the x axis

 
1 0 0
Rx (θ) = 0 cos θ − sin θ
0 sin θ cos θ
• Rotation an angle θ about the y axis
 
cos θ 0 sin θ
Ry (θ) =  0 1 0 
− sin θ 0 cos θ
• Rotation an angle θ about the z axis
 
cos θ − sin θ 0
Rz (θ) =  sin θ cos θ 0 .
0 0 1
Matrices can be found for rotation of any angle about any axis, where an axis is
given by a non-zero vector.
10.11. Problems.
(1) Read about finding eigenvalues using Maple. Let

 
−4 −3 3 −3
 3 2 −3 3
 

C= 
 3
 3 −4 3 

6 6 −6 5
Use Maple to find a non-zero vector X such that CX = 2X.
(2) Use Maple to find the inverses of the matrices
 
1 2 3
2 2−i
M = 3 4 5 and N =
  .
2 + i −2
3 5 6
(3) Suppose A is self adjoint and that v1 and v2 are eigenvectors corresponding
to distinct eigenvalues. Show that v1∗ v2 = 0.
(4) If A is a real 2 × 2 matrix such that A0 A = I and det A = 1, show that for
some θ,
cos θ − sin θ
A= .
sin θ cos θ
(5) Let Rθ be the rotation matrix

cos θ − sin θ
.
sin θ cos θ
Show that
(Rθ − I) (R−θ − I) = 2(1 − cos θ)I.
(6) Show that if A is real and has real eigenvalue λ, then there is a real vector
(a vector with real coordinates) which is an eigenvector corresponding to
λ.
86 J. R. QUINE
(7) Let

0 1 0 −i 1 0
Σ1 = Σ2 = Σ3 = .
1 0 i 0 0 −1
Use Maple to verify the following identities:
Σ21 = 1, Σ22 = 1 Σ23 = 1,
Σ1 Σ2 = iΣ3 , Σ2 Σ1 = −iΣ3 ,
Σ2 Σ3 = iΣ1 , Σ3 Σ2 = −iΣ1 ,
Σ3 Σ1 = iΣ2 , Σ1 Σ3 = −iΣ2 .
(8) For the each of the matrices Σk in the previous problem, find vectors X
and Y such that
Σk X = X and Σk Y = −Y.

Mathematical Techniques in Structural Biology - J. R. Quine

Uploaded by

Copyright:

Available Formats

Mathematical Techniques in Structural Biology - J. R. Quine

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mathematical Techniques in Structural Biology - J. R. Quine

Uploaded by

Copyright:

Available Formats

5.6. Ramachandran diagram 36

Protein structures are stored in a database called the Research Collaboratory

1. Molecular Genetics: DNA

1.1. Genetic code. DNA is a long molecule consisting of a string of base-pairs.

Figure 1. DNA basepairs showing the geometry of the hydrogen bonds

Figure 2. DNA structure

Figure 3. DNA helix indicating 10 basepairs per 360 degree turn

Figure 4. Rosalind Franklin’s x-ray diffraction picture of DNA

2. Molecular Genetics: Proteins

Figure 5. Figure from Introduction to Protein Architecture, by

2.3. Amino acid template. We are interested in the 3D structure of proteins.

Figure 6. Template for amino acid. R denotes the side chain,

Figure 7. Side chain topology

To get an idea of the geometry of a tetrahedron, a regular four-sided solid, you

cos θ = A · B/|A||B| = −1/3.

Compute that θ = arccos(−1/3) ≈ 109.47◦ .

Figure 8. Structure of leucine shown in a computer graphics stick

Figure 9. Peptide Bond. When two amino acids bond together

Figure 10. Peptide plane geometry. a) shows the distribution of

2.8. Secondary structure. The organization of the atoms of a protein is complex

Figure 11. Painting by Irving Geis of a stick model of sperm

Figure 12. Stick diagram of the backbone of an alpha helix, show-

Figure 13. Diagram showing as dotted lines hydrogen bonds be-

Figure 14. Ribbon diagram for the protein myoglobin. The

Figure 15. Part of the protein carboxopeptidase A containing

3. Frames and moving frames

3.1. Basic definitions. A sequence of three vectors v1 , v2 , v3 can be made into

If the determinant is positive, the frame is said to be right-handed, and if the

3.2. Frames and gram matrices. If F = (v1 , v2 , v3 ) is a matrix, the entry in

This is equivalent to the gram matrix being the identity,

If F satisfies this condition it is said to be orthogonal. A frame that is not orthogonal

Figure 16. A moving frame (Frenet frame).

The Frenet frame for the helix is computed by (4) to be

using the product rule to get

4. Orthogonal transformations and Rotations

(12) A(v × w) = Av × Aw,

Figure 18. The geometric definition of the cross product. The

4.1.2. Two dimensions. In two dimensions, every rotation is of the form

See section 10 problem 4.

4.2. Complex form of a rotation. In dimension 2 it is convenient to use complex

where λ = eiθ . The determinant of A is the product of the eigenvalues, so for

(21) R (u, θ) = R (−u, −θ)

(25) AR (u, θ) A−1 = R (Au, θ)

(26) R (u, θ) v = cos θ (v − (u · v)u)

5. Torsion angles and pdb files

Figure 19. Torsion angle φ = Tor (p1 , p2 , p3 , p4 ). The angle is

the centers of the atoms. Let

FIG. 3. Bond vectors, bond angles, and the dihedral angle.

Figure 20. the The

In Maple the iscommand

Figure 21. Naming convention for torsion angles along the

ATOM 1562 CA PHE A 206 -11.262 18.905 53.109 1.00 23.61 C

Figure 22. The Ramachandran plot giving the number of

Diamond packing = fcc ∪ {fcc + (1/2, 1/2, 1/2)} .

6. X-ray Crystallography, Lattices, and Fourier Series

• a high quality crystal is formed from a sample of protein

6.1. Lattices. The basic structure of a crystal is that of a lattice. A crystal is

A lattice is easy to describe mathematically. A lattice L in three dimensions

For an illustration of lattices see the Maple demo.