PCA Biology

Download as ppt, pdf, or txt
Download as ppt, pdf, or txt
You are on page 1of 45

Principal Component Analysis (PCA)

Data Reduction
summarization of data with many (p) variables by a smaller set of (k) derived (synthetic, composite) variables.
p k

Data Reduction
Residual variation is information in A that is not retained in X balancing act between
clarity of representation, ease of understanding oversimplification: loss of important or relevant information.

probably the most widely-used and wellknown of the standard multivariate methods invented by Pearson (1901) and Hotelling (1933)
first applied in ecology by Goodall (1954) under the name factor analysis (principal factor analysis is a synonym of PCA).

Principal Component Analysis (PCA)

takes a data matrix of n objects by p variables, which may be correlated, and summarizes it by uncorrelated axes (principal components or principal axes) that are linear combinations of the original p variables the first k components display as much as possible of the variation among objects.

Principal Component Analysis (PCA)

objects are represented as a cloud of n points in a multidimensional space with an axis for each of the p variables the centroid of the points is defined by the mean of each variable

Geometric Rationale of PCA

the variance of each variable is the average squared deviation of its n values around the mean of that variable. n

1 2 X im X i Vi n 1 m 1

Geometric Rationale of PCA


degree to which the variables are linearly correlated is represented by their covariances.

1 n X im X i X jm X j Cij n 1 m 1
Covariance of variables i and j Sum over all n objects Value of Mean of variable i variable i in object m Value of variable j in object m Mean of variable j

Geometric Rationale of PCA


objective of PCA is to rigidly rotate the axes of this p-dimensional space to new positions (principal axes) that have the following properties:
ordered such that principal axis 1 has the highest variance, axis 2 has the next highest variance, .... , and axis p has the lowest variance covariance among each pair of the principal axes is zero (the principal axes are uncorrelated).

variables X1 and X2 have positive covariance & each has a similar variance.
14 12

2D Example of PCA

10

Variable X 2

X 2 4.91

0 0 2 4 6 8

X 1 8.35
10 12 14 16 18 20

Variable X1

V1 6.67

V2 6.24

C1, 2 3.42

Configuration is Centered
each variable is adjusted to a mean of zero (by subtracting the mean from each value).
8 6

Variable X 2

0 -8 -6 -4 -2 0 2 4 6 8 10 12

-2

-4

-6

Variable X1

PC 1 has the highest possible variance (9.88) PC 2 has a variance of 3.03 PC 1 and PC 2 have zero covariance.
6 4

Principal Components are Computed

PC 2

0 -8 -6 -4 -2 0 2 4 6 8 10 12

-2

-4

-6

PC 1

The Dissimilarity Measure Used in PCA is Euclidean Distance


PCA uses Euclidean Distance calculated from the p variables as the measure of dissimilarity among the n objects PCA derives the best possible k dimensional (k < p) representation of the Euclidean distances among objects.

In practice nobody uses PCA with only 2 variables

Generalization to p-dimensions

The algebra for finding principal axes readily generalizes to p variables


PC 1 is the direction of maximum variance in the p-dimensional cloud of points PC 2 is in the direction of the next highest variance, subject to the constraint that it has zero covariance with PC 1.

Generalization to p-dimensions
PC 3 is in the direction of the next highest variance, subject to the constraint that it has zero covariance with both PC 1 and PC 2 and so on... up to PC p

each principal axis is a linear combination of the original two variables PCj = ai1Y1 + ai2Y2 + ainYn aijs are the coefficients for factor i, multiplied by the measured value for variable j
8

PC 1
6

PC 2
Variable X 2
2

0 -8 -6 -4 -2 0 2 4 6 8 10 12

-2

-4

-6

Variable X1

PC axes are a rigid rotation of the original variables PC 1 is simultaneously the direction of maximum variance and a least-squares line of best fit (squared distances of points away from PC 1 are minimized).
8

PC 1
6

PC 2
Variable X 2
2

0 -8 -6 -4 -2 0 2 4 6 8 10 12

-2

-4

-6

Variable X1

Generalization to p-dimensions
if we take the first k principal components, they define the k-dimensional hyperplane of best fit to the point cloud of the total variance of all p variables:
PCs 1 to k represent the maximum possible proportion of that variance that can be displayed in k dimensions i.e. the squared Euclidean distances among points calculated from their coordinates on PCs 1 to k are the best possible representation of their squared Euclidean distances in the full p dimensions.

Covariance vs Correlation
using covariances among variables only makes sense if they are measured in the same units even then, variables with high variances will dominate the principal components these problems are generally avoided by standardizing each variable to unit variance and zero mean.

X im

im

Xi Standard deviation of variable i SDi


Mean variable i

Covariance vs Correlation
covariances between the standardized variables are correlations

after standardization, each variable has a variance of 1.000 correlations can be also calculated from the variances and covariances:
Correlation between variables i and j

rij

C ij ViV j

Covariance of variables i and j Variance of variable j

Variance of variable i

first step is to calculate the crossproducts matrix of variances and covariances (or correlations) among every pair of the p variables square, symmetric matrix

The Algebra of PCA

diagonals are the variances, off-diagonals are the covariances.


X1
X1 X2 6.6707 3.4170

X2
3.4170 6.2384 X1 X2

X1 1.0000 0.5297 Correlation Matrix

X2 0.5297 1.0000

Variance-covariance Matrix

in matrix notation, this is computed as

The Algebra of PCA

where X is the n x p data matrix, with each variable centered (also standardized by if using correlations).
X1
X1 X2 6.6707 3.4170

S XX
X2
3.4170 6.2384 X1 X2

SD

X1 1.0000 0.5297 Correlation Matrix

X2 0.5297 1.0000

Variance-covariance Matrix

Manipulating Matrices
transposing: could change the columns to rows or the rows to columns
X = 10 0 4 7 1 2 X = 10 7 0 1 4 2

multiplying matrices
must have the same number of columns in the premultiplicand matrix as the number of rows in the postmultiplicand matrix

sum of the diagonals of the variancecovariance matrix is called the trace it represents the total variance in the data it is the mean squared Euclidean distance between each object and the centroid in p-dimensional space.
X1
X1 X2 6.6707 3.4170

The Algebra of PCA

X2
3.4170 6.2384 X1 X2

X1 1.0000 0.5297 Trace = 2.0000

X2 0.5297 1.0000

Trace = 12.9091

finding the principal axes involves eigenanalysis of the cross-products matrix (S) the eigenvalues (latent roots) of S are solutions () to the characteristic equation

The Algebra of PCA

S I 0

the eigenvalues, 1, 2, ... p are the variances of the coordinates on each principal component axis the sum of all p eigenvalues equals the trace of S (the sum of the variances of the original variables).
X1 X1 X2 6.6707 3.4170 X2 3.4170 6.2384

The Algebra of PCA

1 = 9.8783 2 = 3.0308
Note: 1+2 =12.9091

Trace = 12.9091

each eigenvector consists of p values which represent the contribution of each variable to the principal component axis
eigenvectors are uncorrelated (orthogonal)
their cross-products are zero.

The Algebra of PCA

Eigenvectors
u1 X1 X2 0.7291 0.6844 u2 -0.6844 0.7291

0.7291*(-0.6844) + 0.6844*0.7291 = 0

zki u1k x1i u2 k x2 i u pk x pi


where Z is the n x k matrix of PC scores, X is the n x p centered data matrix and U is the p x k matrix of eigenvectors.

coordinates of each object i on the kth principal axis, known as the scores on PC k, are computed as

The Algebra of PCA

variance of the scores on each PC axis is equal to the corresponding eigenvalue for that axis the eigenvalue represents the variance displayed (explained or extracted) by the kth axis

The Algebra of PCA

the sum of the first k eigenvalues is the variance explained by the k-dimensional ordination.

1 = 9.8783

2 = 3.0308

Trace = 12.9091

PC 1 displays (explains) 9.8783/12.9091 = 76.5% of the total variance


6

PC 2

0 -8 -6 -4 -2 0 2 4 6 8 10 12

-2

-4

-6

PC 1

The Algebra of PCA


The cross-products matrix computed among the p principal axes has a simple form:
all off-diagonal values are zero (the principal axes are uncorrelated) the diagonal values are the eigenvalues.
PC1 PC1 PC2 9.8783 0.0000 PC2 0.0000 3.0308

Variance-covariance Matrix of the PC axes

data from research on habitat definition in the endangered Baw Baw frog 16 environmental and structural variables measured at each of 124 sites correlation matrix used because variables have different units

A more challenging example

Philoria frosti

Eigenvalues
Axis 1 2 3 4 5 6 7 8 9 10 Eigenvalue 5.855 3.420 1.122 1.116 0.982 0.725 0.563 0.529 0.476 0.375 % of Variance 36.60 21.38 7.01 6.97 6.14 4.53 3.52 3.31 2.98 2.35 Cumulative % of Variance 36.60 57.97 64.98 71.95 78.09 82.62 86.14 89.45 92.42 94.77

Interpreting Eigenvectors
correlations between variables and the principal axes are known as loadings each element of the eigenvectors represents the contribution of a given variable to a component
1 2 Altitude pH Cond TempSurf Relief maxERht avERht %ER %VEG %LIT %LOG %W H1Moss DistSWH DistSW DistMF 0.3842 -0.1159 -0.2729 0.0538 -0.0765 0.0248 0.0599 0.0789 0.3305 -0.3053 -0.3144 -0.0886 0.1364 -0.3787 -0.3494 0.3899 0.0659 0.1696 -0.1200 -0.2800 0.3855 0.4879 0.4568 0.4223 -0.2087 0.1226 0.0402 -0.0654 -0.1262 0.0101 -0.1283 0.0586

3 -0.1177 -0.5578 0.3636 0.2621 -0.1462 0.2426 0.2497 0.2278 -0.0276 0.1145 -0.1067 -0.1171 0.4761 0.0042 0.1166 -0.0175

How many axes are needed?


does the (k+1)th principal axis represent more variance than would be expected by chance? several tests and rules have been proposed a common rule of thumb when PCA is based on correlations is that axes with eigenvalues > 1 are worth interpreting

Baw Baw Frog - PCA of 16 Habitat Variables


7.0 6.0 5.0

Eigenvalue

4.0 3.0 2.0 1.0 0.0 1 2 3 4 5 6 7 8 9 10

PC Axis Number

What are the assumptions of PCA?


assumes relationships among variables are LINEAR
cloud of points in p-dimensional space has linear dimensions that can be effectively summarized by the principal axes

if the structure in the data is NONLINEAR (the cloud of points twists and curves its way through p-dimensional space), the principal axes will not be an efficient and informative summary of the data.

In community ecology, PCA is useful for summarizing variables whose relationships are approximately linear or at least monotonic
e.g. A PCA of many soil properties might be used to extract a few components that summarize main dimensions of soil variation

When should PCA be used?

PCA is generally NOT useful for ordinating community data

Why? Because relationships among species are highly nonlinear.

70

60

50

Abundance

40

30

20

10

0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Simulated Environmetal Gradient (R)

The Horseshoe or Arch Effect


community trends along environmenal gradients appear as horseshoes in PCA ordinations none of the PC axes effectively summarizes the trend in species composition along the gradient SUs at opposite extremes of the gradient appear relatively close together.

Ambiguity of Absence

Abundance

Environmental Gradient

Beta Diversity 2R - Covariance

Axis 2

Axis 1

The HorseshoeEffect
curvature of the gradient and the degree of infolding of the extremes increase with beta diversity PCA ordinations are not useful summaries of community data except when beta diversity is very low using correlation generally does better than covariance
this is because standardization by species improves the correlation between Euclidean distance and environmental distance.

What if theres more than one underlying ecological gradient?

The Horseshoe Effect


when two or more underlying gradients with high beta diversity a horseshoe is usually not detectable the SUs fall on a curved hypersurface that twists and turns through the pdimensional species space interpretation problems are more severe

PCA should NOT be used with community data (except maybe when beta diversity is very low).

Impact on Ordination History


by 1970 PCA was the ordination method of choice for community data
simulation studies by Swan (1970) & Austin & Noy-Meir (1971) demonstrated the horseshoe effect and showed that the linear assumption of PCA was not compatible with the nonlinear structure of community data

stimulated the quest for more appropriate ordination methods.

You might also like