See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/10584372

Bioinformatic strategies for better understanding of immune function

Article  in  In Silico Biology · February 2003

CITATIONS READS

18 224

3 authors:

Nikolai Petrovsky Christian Schönbach

SEE PROFILE SEE PROFILE

Vladimir Brusic
University of Nottingham Ningbo China
339 PUBLICATIONS   15,125 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Elemental Metabolomics View project

Cell classification with single-cell transcriptomics and supervised machine learning View project

All content following this page was uploaded by Vladimir Brusic on 27 February 2018.

The user has requested enhancement of the downloaded file.

Special Feature
Computational immunology: The coming of age
NIKOLAI PETROVSKY1 and VLADIMIR BRUSIC2
1National Bioinformatics Centre, University of Canberra and National Health Sciences Centre, Canberra Clinical
School, Woden, Australian Capital Territory, Australia and 2Kent Ridge Digital Laboratories, Singapore

Summary The explosive growth in biotechnology combined with major advances in information technology has
the potential to radically transform immunology in the postgenomics era. Not only do we now have ready access to
vast quantities of existing data, but new data with relevance to immunology are being accumulated at an exponential
rate. Resources for computational immunology include biological databases and methods for data extraction,
comparison, analysis and interpretation. Publicly accessible biological databases of relevance to immunologists
number in the hundreds and are growing daily. The ability to efficiently extract and analyse information from these
databases is vital for efficient immunology research. Most importantly, a new generation of computational
immunology tools enables modelling of peptide transport by the transporter associated with antigen processing
(TAP), modelling of antibody binding sites, identification of allergenic motifs and modelling of T-cell receptor
serial triggering.

Key words: allergy, autoimmunity, bioinformatics, cancer, computational immunology, databases, genomics.

Introduction Databases
The difficulties in the development and application of compu- There has been an explosion in general-purpose biological
tational immunology tools include the high complexity of databases that contain annotated biological sequence entries. 1
immunological processes, the intrinsic imprecision of bio- Most commonly, these databases contain the actual sequence,
logical measurements, and the biases and misconceptions a short description, the name and the taxonomy of the source
inherent to the scientific endeavour. Care needs to be taken organism, and in some cases tables listing features and
that computational tools are adequate and that they properly bibliographic data. Amongst the best-known generic data-
model the underlying immunological processes. Biases and bases are the nucleic acid databases, GenBank 2 and European
misconceptions can easily be encoded in computational tools, Molecular Biology Laboratory (EMBL), 3 and the protein
resulting in the misinterpretation of insights provided by databases, SWISS-PROT4 and Protein Information Resource
computational approaches. One role of computational immu- (PIR).5 Table 1 lists the URL addresses for a number of these
nology is to convert immunological into computational prob- generic databases. Information on protein families and
lems, solve computational problems using mathematical and domains, including information on biologically significant
computational approaches and convert these results into bio- sites, patterns and profiles, can be accessed through the
logically meaningful interpretations. PROSITE database.6 This contains more than 10 000 3-D
Bioinformatic tools available for database searching and macromolecular structures, including several hundred struc-
biological sequence analysis have become increasingly tures of immune-system products.
sophisticated. These tools enable the quick identification of Although the generic databases listed in Table 1 contain
sequences of interest and provide some bibliographic, taxo- large numbers of sequences of immunological relevance, the
nomic and feature information. Tools for sequence compari- annotation of the sequences in these databases is often limited
son, motif searching or sequence profiling assist researchers to basic facts. Fortunately, more specific databases also exist
to identify biologically relevant sequence similarities and that contain immunologically relevant sequences. A compre-
features. A newer generation of computational tools enables hensive list of databases is available through the Molecular
the modelling of biological interactions and simulation of Biology Database Collection,7 or online through the Public
laboratory experiments. These tools help researchers identify Catalogue of Databases (DBCAT) at http://www.infobiogen.fr/
and focus upon relevant experiments, thus speeding up the services/dbcat/file/dbcat.html. This site currently (January
discovery process. Therefore, biological databases, sequence 2002) lists 87 DNA, 29 RNA, 94 protein, 58 genomic, 29
comparison tools, and more complex modelling and simula- mapping, 18 protein structure, 43 literature and 153 miscella-
tion tools, are core resources for computer-assisted discovery neous databases.
and data analysis. Searching generic sequence databases is an essential tool
Correspondence: Associate Prof. N Petrovsky, National Health in many immunology projects, providing useful hits or
Sciences Centre, Autoimmunity Research Unit, Canberra Hospital, insights into a sequence of interest. When more sophisticated
PO Box 11, Woden, ACT 2606, Australia. analyses are required (such as elucidation of related biological
Email: [email protected] pathways or molecular function amongst others), more spe-
Received 25 February 2002; accepted 25 February 2002. cific and detailed databases may be required. In many cases,
 Computational immunology 249

Table 1 A short list of the best-known, general-purpose sequence MHCPEP (http://wehih.wehi.edu.au/mhcpep/)10 is a data-
databases base comprising over 13 000 peptide sequences known to
bind MHC molecules. Entries were compiled from published
Database URL reports as well as from direct submissions of experimental
DBCAT Catalogue of http://www.infobiogen.fr/services/dbcat data. Each entry contains the peptide sequence, its MHC
Databases specificity and, when available, experimental method, observed
GenBank http://www.ncbi.nlm.nih.gov activity, binding affinity, source protein, anchor positions and
EMBL http://www.ebi.ac.uk/embl publication references.
Protein Information http://www-nbrf.georgetown.edu/pir
FIMM is an annotated database of functional immuno-
Resource (PIR)
SWISS-PROT http://www.expasy.ch/sprot
logy.11 FIMM includes data on MHC, protein antigens, anti-
PROSITE http://www.expasy.ch/prosite genic peptides and relevant disease information. Tools
PDB http://www.rcsb.org/pdb include searches by key word, sequence pattern and basic
local alignment search tool (BLAST), multiple sequence
EMBL, European Molecular Biology Laboratory; PDB, Protein alignment and binding pocket/contact sites analysis.
data bank. The SYFPEITHI database (http://syfpeithi.bmi-heidelberg.
com/scripts/MHCServer.dll/home.htm) contains data on peptide
these specialty databases provide better overall quality of sequences, anchor positions, MHC specificity and tools for
information, at least partly because they are annotated by epitope prediction.12
experts from that particular field. This raises the commonly FARRP (http://www.allergenonline.com/default.asp) con-
neglected issue of database quality that has major relevance tains a list of publicly known allergens. The National Center
for casual users of both generic and specialty databases alike for Food Safety and Technology has created three databases
and will be dealt with later in this review. of known allergen sequences in order to facilitate the use of
sequence comparisons in assessing the potential allergenicity
of food proteins (http://www.iit.edu/∼sgendel/fa.htm). These
Specialist immunology databases
databases contain accession numbers for food allergens, non-
An example of a specialist immunology database is the HIV food allergens and wheat gluten proteins.
Molecular Immunology Database (http://hiv-web.lanl.gov/ The Protall Database (http://www.ifrn.bbsrc.ac.uk/protall/
immunology/). This contains T-cell epitope maps on HIV pro- database.html) lists biochemical and clinical information
teins, alignments and annotations, and links to the HIV about plant food allergens involved in classical IgE-induced
Sequence Database (http://hiv-web.lanl.gov/) that contains hypersensitivity reactions.
HIV genetic sequences, immunological epitopes, drug- Prediction Algorithm for Proteasomal Cleavages (PAProC;
resistance-associated mutations and vaccine trials. Annotation http://www.paproc.de), whilst not strictly a database, is a
includes information such as how specific epitopes were potentially useful immunological tool. PAProC predicts
experimentally defined, HLA specificities, isotypes of mAb, cleavage by human and yeast proteasomes, based on experi-
the initial immunogen, and brief notes describing the context mental cleavage data and is thus useful for immunologists
in which a given epitope was studied. Associated database working on antigen processing and the prediction of MHC
tools include Epilign (aligns amino acid epitopes against class I molecule ligands and CTL epitopes. 13
alignments), PeptGen (maps overlapping peptides, with BioInformatics & Molecular Analysis Section (BIMAS;
options to adjust length, overlap, and to exclude user-selected http://bimas.dcrt.nih.gov/) is a site designed to provide guid-
amino acids from C- or N-terminal positions), Motifscan ance, support and resources to scientists throughout the
(scans sequences for possible epitopes based on HLA-binding National Institutes of Health in bioinformatics, genomics and
motifs), Primalign (aligns nucleotide sequences against align- genetic analysis. Examples of BIMAS tools accessible over
ments), the HXB2 numbering engine, SeqPublish and BLAST. the web include an HLA-binding prediction program, promoter
The International ImMunoGeneTics database (IMGT; scan (the PROSCAN program), a signal scan (SIGSCAN),
http://imgt.cines.fr:8104/)8 is a high-quality integrated data- sequence format conversion program ( READSEQ) and identifi-
base specializing in immunoglobulins, T-cell receptors, and cation of tissue-enriched (human or mouse) unigene clusters.
MHC of all vertebrate species. At present, IMGT includes The specialist immunology databases are becoming
two databases: (i) LIGM-DB, a comprehensive database of increasingly numerous and sophisticated. The functionality of
immunoglobulins and T-cell receptors from human and other databases will increase as more high-level analysis tools are
vertebrates, with translation for fully annotated sequences; integrated into databases. However, this additional complex-
and (ii) HLA-DB, a database of HLA. ity will also require increasingly sophisticated and knowl-
KABAT (http://immuno.bme.nwu.edu/),9 the oldest spe- edgeable users if traps and erroneous results are to be
cialist immunology database, includes nucleotide sequences, avoided. Some of the potential traps arising from database
sequences of TCR, MHC class I and II molecules, and other design and usage are described below.
proteins of immunological interest. It includes tools such as
SeqhuntII, Variability, Align-A-Sequence, Subgrouping and
Find Your Family. Database issues
The Swine Major Histocompatibility Complex (SLA)
database (SLAD; http://sdmc.krdl.org.sg/bic/projects/p19.html)
Data standardization
provides the basis for establishing SLA nomenclature through Standardization and nomenclature are important so as to
phylogenetic analyses. allow a consistent view of the data available in a particular
250 N Petrovsky and V Brusic

field. Databases play an important role because of the need to research efficiency, immunologists require both efficient
unify existing scientific classification and nomenclature with database facilities to manage and store complex information
data structures suitable for analysis by computers. An impor- plus relevant data, and sequence analysis tools to perform
tant objective of the IMGT database is to contribute to the complex analyses on extracted data. Hundreds of programs
standardization of data in immunology. Amongst other uses, for biological data analysis have been developed in recent
standardized nomenclature enables cross-species compari- years. A listing of more than 130 of these bioinformatics tools
sons, thereby providing avenues to gain new insights into is accessible via eBioinformatics BioNavigator at http://
molecular structure/function relationships. www.bionavigator.com. The FIMM database,11 for example,
integrates BLAST and pattern searches, multiple sequence
Data interpretation analysis and binding site analysis in the database interface.
Databases that integrate tools enabling complex data analysis
Database developers are well aware of the difficulties in will be increasingly indispensable to biomedical research.
building reliable and effective databases. However, if users Although individual databases are often readily accessible, it
are to efficiently use these resources then they also need a is often difficult to extract information from heterogeneous
good understanding of relevant issues. The interpretation of databases because of a lack of standard nomenclature and
data extracted from databases is highly dependent on the variable formats. This problem is slowly being overcome and
skills and knowledge of the user. Lack of standards, ad hoc systems such as Kleisli16 or CORBA17 provide means for
nomenclature, variable source data quality, incomplete infor- efficient extraction of data from heterogeneous sources.
mation and biases are all major obstacles to data interpreta- These systems facilitate the automation of large-scale data
tion. Furthermore, most database search tools require careful access, extraction and integration.
selection of parameters for search optimization, as default
settings are often not optimal for a specific query. The
Specific applications of computational immunology
selection of search parameters requires a good understanding
of the specific issues associated with a particular query. The Allergies
development of specialist databases enhances the quality and
The number of known protein allergens is increasing rapidly,
availability of data, but human experts are still necessary for
with some 360 protein allergens having been classified as of
annotation and interpretation.
September 2001. The assessment of protein allergenic poten-
tial focuses on three main aspects: (i) immunogenicity; (ii)
Data quality cross-reactivity; and (iii) clinical symptoms. 18 Immunogenic-
Database quality is a descriptive measure of the extent to ity reflects the likelihood of an IgE antibody and/or T-cell
which a database is capable of achieving its intended purpose. response to a particular allergen. Therefore, immunogenicity
As researchers and end-users, we would like to think that all studies focus on identifying B-cell19,20 and T-cell21,22 recogni-
the data held within a database is of uniformly high quality. tion sites on allergens. The 3-D structural properties of
Unfortunately, this is rarely the case. In order to properly allergens control their allergenicity, and four families of
assess database quality we need to address its individual allergens have been defined recently based on 3-D structure. 23
components. These include data accuracy and accessibility, Allergen-related data are currently spread across a number of
ease of use, breath and depth of coverage, ease of integration publicly accessible molecular databases, including the major
with data from other sources, and performance of data analy- protein databases (PIR, SwissProt and TrEMBL). As these
sis tools attached to the database. Errors in public databases databases do not contain detailed structural and functional
can result from biological variation, experimental artefacts, allergen information, various specialized allergy-related data-
imperfect sequence determination and poor data handling, bases and data repositories have been developed. Bioinfor-
amongst others. For example, the GenBank database included matic prediction tools enabling the identification of proteins
0.36% vector-contaminated sequences in release 1995–1996 with allergenic potential are available for prediction of aller-
and had a 10–20% rate of erroneous annotation of entries as genicity, allergen cross-reactivity,24 and T-cell epitopes.22
‘genes’.14 Although current standards call for genomic Structural bioinformatic tools help identify structural proper-
sequences to have an error rate of less that one error per ties of proteins, such as secondary structure, 25 or tertiary
10000 bases,15 the sequence error rate is much higher than structure,26,27 which affect allergenicity through affecting IgE
this for older data. This has been compounded by the use of binding sites. Through the collation and analysis of the
automatic annotation of database entries, which, although protein sequences of known allergens and their epitopes, it is
faster, tends to expand existing errors and introduce new possible to identify related groups that correlate with
errors. Overall error rates in functional assignment may be as observed clinical cross-reactivities. Comparative molecular
high as 2.5–5.0%. What this demonstrates is the need for modelling was successful in identifying a common putative
careful annotation, including human expert assessment of IgE epitope on cysteine protease allergens of diverse sources. 23
each database entry. Examples of expert-annotated immuno- The use of bioinformatic tools to predict protein aller-
logical databases include the KABAT, 9 IMGT8 and FIMM11 genicity will become increasingly important in the screening
databases. of novel foods before their wide-scale release for human
use.18 Thus, there is a major need to develop reliable broad-
based allergy databases and combine these with well-
Database tools
validated bioinformatic prediction tools in order to enable the
Databases require tools to enable efficient extraction of data identification of potential allergens in genetically modified
and generation of knowledge. Therefore, to increase their drugs and foods.
 Computational immunology 251

Infectious diseases and host responses HLA class I antigen presentation.56 TAP is a transmembrane
protein responsible for the transport of antigenic peptides into
Mathematical and computer modelling can be of great help in
the endoplasmic reticulum, where they can be bound by MHC
understanding the behaviour and spread of infectious disease.
class I molecules and presented to T cells. As TAP does not
Examples include mathematical modelling of the within-host
bind all peptides equally, TAP-binding affinity could influ-
dynamics of Plasmodium falciparum,19 determination of host
ence the ability of a particular peptide to gain access to the
factors influencing viral persistence, 20 development of a
MHC class I pathway. However, analysis of the relationship
stochastic model of schistosomiasis immuno-epidemiology, 21
between TAP- and MHC class I-binding has been hampered
and of nematode infection in ruminants, 22 modelling the
by the extensive polymorphism of MHC molecules (>200
dynamics of lymphocytic choriomeningitis virus (LCMV)
HLA class I allelic variants exist). To overcome these prob-
infection in mice,28 and modelling the likelihood of rubella
lems, we used an artificial neural network (ANN) computer
epidemics in Europe.29 Computer simulations have been per-
model to study peptide binding to human TAP and its
formed on the transmission dynamics and the effects of
relationship with MHC class I binding. 56 After development
duration of immunity and survival of persistently infected
and validation, the ANN model of TAP-peptide binding was
animals in the spread of bovine viral diarrhoea virus in dairy
used to mine a database of HLA-binding peptides to elucidate
cattle.30 Trypanosoma congolense parasitaemia patterns have
patterns of TAP binding. The affinity of HLA-binding pep-
been modelled during the chronic phase of infection in
tides for TAP was found to differ according to the HLA
N’Dama cattle.31 In addition, there has been development of a
supertype concerned: HLA-B27-, -A3- or -A24-binding pep-
theoretical framework for the immunoepidemiology of Plas-
tides had high affinities for TAP, whereas HLA-A2-, -B7- or
modium falciparum malaria,32 and of the development resist-
-B8-binding peptides had low affinities for TAP. The affini-
ance of P. falciparum to antimalarial drugs.33 More recently,
ties of HLA-A2- and HLA-B7-binding peptides for TAP were
extensive work has been conducted on modelling HIV vacci-
shown to have similar distributions to that of randomly
nation,34 and HIV-infection therapy.35
generated peptides. This work showed for the first time that
A modelling approach has also been used to better under-
HLA molecules constitute two separate classes: (i) those that
stand control of infections by immunization. This has been
are TAP-dependent for peptide loading (HLA-B27, -A3 and
studied in respect of rabies control in fox populations, 36 the
-A24); and (ii) those that are relatively TAP-independent
modelling of antibody response to measles vaccine and subse-
(HLA-A2, -B7 and -B8). The strong similarity between the
quent waning of immunity in a low exposure population, 37 the
sets of peptides bound by TAP and HLA-B27, -A3 or -A24
impact of immunization on the epidemiology of Varicella
suggest the possibility of their functional coevolution. This
zoster virus,38 and the impact of mass vaccination against
could have important implications for the design of peptide-
hepatitis B.39 Mathematical models have been used to assess
based immuno-therapeutic drugs and vaccines, and shows the
the long-term persistence of antibodies after vaccination with
power of the modelling approach in helping to better under-
two inactivated hepatitis A vaccines,40,41 and to identify
stand complex immune interactions.
subcritical endemic steady states for animal infections with
incomplete immunity.24
These examples illustrate the power of a computational
approach to complex problems involving potentially vast Antibody–antigen–MHC interactions
datasets of sometimes imprecise clinical data. The develop-
ment of computer models of infectious diseases and their Various models have been developed to understand antibody–
response to treatment or vaccination allows the initial testing antigen interactions. Modelling of ferritin antibody–antigen
of hypotheses, which can then be validated in the field or the interactions based on experimental data helped explain the
laboratory. This offers the ability to speed up the discovery inability of the antibody to recognize human L-ferritin and
process dramatically and to potentially shorten the time for rat, chicken and mouse H-ferritin.57 Models have been used to
vaccine development for major infections such as HIV. study IgE/receptor interactions,58 and the MHC class II allele
I-A(g7) of NOD mice.59 A number of studies have shown the
power of combining experimental results with a modelling
Immune system function
approach to understand antibody–antigen interactions. 57,60
Modelling has also been extremely helpful in the understand- Other examples of the use of immune interaction modelling
ing of immune population dynamics. It has been used to include the prediction of epitopes and the production of mAb
model T-cell-mediated suppression,25 peripheral lymphocyte against gastric H,K-ATPase,61 modelling of the 3-D structure
migration,26 T-cell memory,27,42 tolerance,43,44 thymic func- for EgDf1 from Echinococcus granulosus,62 modelling the
tion,45,46 and antibody networks.47 Models predict the dynam- kinetics of antigen–antibody reactions in particle-enhanced
ics of antiviral cytotoxic T-cell memory in response to optical immunoassays,63 the construction of an antibody
different stimuli.48 Shape–space modelling has been success- against cystatin,64 the use of cellular automaton to study local
fully used to understand the nature of specificity in an T–T-cell and T–B-cell interactions,65 and to demonstrate the
immune network.49 Modelling has been particularly success- importance of efficacy and partial agonism in models of
ful in predicting TCR serial triggering and downregulation, 50 B-lymphocyte activation.66
HLA binding,51,52 T-cell epitopes,53 and immunogenicity.54,55 These examples demonstrate the power of modelling
A demonstration of the power of a computational model- approaches in the understanding of the interaction between
ling approach to the understanding of immune system func- complex 3-D surfaces, such as the binding of antibody to
tion is shown by a recent study where we sought to examine antigen. In addition, computational approaches are also well
the function relationship between TAP peptide transport and suited to numerical problems, such as understanding the
252 N Petrovsky and V Brusic

multiplicity of antibody and T-cell idiotypes and how these for predicting both the overall outcomes of renal transplants
all come together to constitute an immune response. as well as the type of rejection. Conceivably, computational
prediction systems have great potential as tools for cadaver
kidney allocation, management of post-transplantation immuno-
Autoimmunity
suppression and the exploration of the role of HLA in
Modelling has mainly been used in the autoimmunity area for transplant rejection.
the prediction of HLA-binding peptides and T-cell epitopes. 53
Other uses have included modelling of MHC susceptibility to
Conclusions
rheumatoid arthritis,67 and identification of antibody epitopes
within the CB-11 peptide of type II collagen. 68 Molecular Computational immunology is a rapidly expanding field that
modelling of HLA-DQ has also been used to suggest a has moved far beyond its origins as a series of databases and
mechanism of resistance in type 1 diabetes. 69 More general search tools. More and more advanced modelling technolo-
models of autoimmunity and its therapy, 70 and the effects of gies are enabling laboratory scientists to better understand
genetic and environmental factors,71 have been developed. their data and to develop new hypotheses to explain immune
Given the basis of autoimmunity being one of the classic system function. The power of this combined approach is
unsolved philosophical problems of modern immunology, the most evident in the close correlation between wet lab experi-
scarcity of computational models that have been developed to mental data and the results obtained using computational
try and resolve this problem is surprising. Similarly, there is a models.
scarcity of databases with a major focus on autoimmunity and
its mechanisms. This is something that we hope to see
References
addressed in the near future.
1 Brusic V, Zeleznikow J, Petrovsky N. Molecular immunology
databases and data repositories. J. Immunol. Meth. 2000; 238:
Tumour immunology 17–28.
Computer models have been used in an attempt to understand 2 Benson DA, Karsch-Mizrachi I, Lipman DJ, Ostell J, Rapp BA,
the processes of adoptive cellular cancer immunotherapy. 72 Wheeler DL. GenBank. Nucleic Acids Res. 2002; 30: 17–20.
Macrophage invasion of tumours and their effects on growth 3 Stoesser G, Baker W, van den Broek A et al. The EMBL nucleo-
tide sequence database. Nucleic Acids Res. 2002; 30: 21–6.
and composition have also been modelled. 73 Predictions have
4 Bairoch A, Apweiler R. The SWISS-PROT protein sequence
been used successfully to identify tumour-derived peptides
database and its supplement TrEMBL in 2000. Nucleic Acids
suitable for use in cancer vaccines.74–76
Res. 2000; 28: 45–8.
This is a relatively new field; therefore, the bioinformatic
5 Wu CH, Huang H, Arminski L et al. The Protein Information
resources for tumour immunology research remain limited. Resource: an integrated public resource of functional annotation
However, given the increasing success of tumour vaccination, of proteins. Nucleic Acids Res. 2002; 30: 35–7.
these resources are likely to grow significantly in the near 6 Falquet L, Pagni M, Bucher P et al. The PROSITE database, its
future. status in 2002. Nucleic Acids Res. 2002; 30: 235–8.
7 Baxevanis AD. The Molecular Biology Database Collection:
Organ transplantation 2002 update. Nucleic Acids Res. 2002; 30: 1–12.
8 Robinson J, Malik A, Parham P, Bodmer JG, Marsh SG. IMGT/
Neural networks have been used to predict the outcome of HLA database – a sequence database for the human major histo-
liver,77,78 pancreas,33 and bone marrow transplantation.79 Other compatibility complex. Tissue Antigens 2000; 55: 280–7.
examples of the use of predictive models in transplantation 9 Johnson G, Wu TT. Kabat Database and its applications: future
include the prediction of tacrolimus blood levels in liver directions. Nucleic Acids Res. 2001; 29: 205–6.
transplantation,30 diagnosis of early acute renal allograft 10 Brusic V, Rudy G, Harrison LC. MHCPEP, a database of MHC-
rejection and evaluation of complications of renal trans- binding peptides: update 1997. Nucleic Acids Res. 1998; 26:
plants,31 prediction of cytomegalovirus disease after renal 368–71.
transplantation,32 and prediction of delayed renal allograft 11 Schonbach C, Koh JL, Flower DR, Wong L, Brusic V. FIMM, a
function.36 database of functional molecular immunology: update 2002.
As an example of how a computational prediction system Nucleic Acids Res. 2002; 30: 226–9.
12 Rammensee H, Bachmann J, Emmerich NP, Bachor OA,
can assist in improving transplantation outcomes, we recently
Stevanovic S. SYFPEITHI database for MHC ligands and
showed that a computational prediction system can accurately
peptide motifs. Immunogenetics 1999; 50: 213–19.
predict graft rejection for particular donor/recipient pairs. 80
13 Nussbaum AK, Kuttler C, Hadeler KP, Rammensee HG, Schild
The prediction system used an ANN trained on renal trans-
H. PAProC: a prediction algorithm for proteasomal cleavages
plant data from the Australian and New Zealand Dialysis and available on the WWW. Immunogenetics. 2001; 53: 87–94.
Transplant Registry (ANZDATA). After training with data 14 Seluja GA, Farmer A, McLeod M, Harger C, Schad PA. Estab-
from 1400 renal transplant recipients, the ANN correctly lishing a method of vector contamination identification in data-
predicted 84.95% of successful and 71.7% of unsuccessful, base sequences. Bioinformatics 1999; 15: 106–10.
transplants. We then tested whether an ANN could predict the 15 Macauley J, Wang H, Goodman N. A model system for studying
type of graft rejection, namely hyper-acute, acute, subacute the integration of molecular biology databases. Bioinformatics
and chronic. The ANN correctly predicted rejection type in 1998; 14: 575–82.
59% of cases compared to a random probability of 25%. 16 Chung SY, Wong L. Kleisli: a new tool for data integration in
These results indicated that an ANN-based approach is useful biology. Trends Biotechnol. 1999; 17: 351–5.
 Computational immunology 253

17 Stevens R, Miller C. Wrapping and interoperating bioinformatics 39 McLean AR, Blumberg BS. Modelling the impact of mass vacci-
resources using CORBA. Brief Bioinform. 2000; 1: 9–21. nation against hepatitis B. I. Model formulation and parameter
18 Oehlschlager S, Reece P, Brown A et al. Food allergy – towards estimation. Proc. R. Soc. Lond. B Biol. Sci. 1994; 256: 7–15.
predictive testing for novel foods. Food Addit. Contam. 2001; 40 Van Herck K, Beutels P, Van Damme P et al. Mathematical
18: 1099–107. models for assessment of long-term persistence of antibodies
19 Hoshen MB, Heinrich R, Stein WD, Ginsburg H. Mathematical after vaccination with two inactivated hepatitis A vaccines. J.
modelling of the within-host dynamics of Plasmodium falci- Med. Virol. 2000; 60: 1–7.
parum. Parasitology 2000; 121: 227–35. 41 Vellinga A, Van Damme P, Bruckers L, Weyler JJ, Molenberghs
20 Thomsen AR, Nansen A, Andreasen SO, Wodarz D, Christensen G, Meheus A. Modelling long-term persistence of hepatitis B
JP. Host factors influencing viral persistence. Philos. Trans. R. antibodies after vaccination. J. Med. Virol. 1999; 57: 100–3.
Soc. Lond. B Biol. Sci. 2000; 355: 1031–41. 42 McLean AR. Modelling T cell memory. J. Theor. Biol. 1994;
21 Chan MS, Isham VS. A stochastic model of schistosomiasis 170: 63–74.
immuno-epidemiology. Math. Biosci. 1998; 151: 179–98. 43 Dolezal J, Hraba T. A contribution to mathematical modelling of
22 Marion G, Renshaw E, Gibson G. Stochastic effects in a model immunological tolerance. Arch. Immunol. Ther. Exp. (Warsz)
of nematode infection in ruminants. IMA J. Math. Appl. Med. 1988; 36: 23–30.
Biol. 1998; 15: 97–116. 44 Borghans JA, de Boer RJ. Neonatal tolerance revisited by mathe-
23 Furmonaviciene R, Sewell HF, Shakib F. Comparative molecu- matical modelling. Scand. J. Immunol. 1998; 48: 283–5.
lar modelling identifies a common putative IgE epitope on 45 Mehr R, Segel L, Sharp A, Globerson A. Colonization of the
cysteine protease allergens of diverse sources. Clin. Exp. Allergy thymus by T cell progenitors: models for cell–cell interactions.
2000; 30: 1307–13. J. Theor. Biol. 1994; 170: 247–57.
24 Greenhalgh D, Diekmann O, de Jong MC. Subcritical endemic 46 Morpurgo D, Serentha R, Seiden PE, Celada F. Modelling
steady states in mathematical models for animal infections with thymic functions in a cellular automaton. Int. Immunol. 1995; 7:
incomplete immunity. Math. Biosci. 2000; 165: 1–25. 505–16.
25 Leon K, Perez R, Lage A, Carneiro J. Modelling T-cell-mediated 47 Faro J, Carneiro J, Velasco S. Further studies on the problem
suppression dependent on interactions in multicellular conju- of immune network modelling. J. Theor. Biol. 1997; 184:
gates. J. Theor. Biol. 2000; 207: 231–54. 405–21.
26 Srikusalanukul W, De Bruyne F, McCullagh P. Modelling of 48 Bocharov G, Klenerman P, Ehl S. Predicting the dynamics of
peripheral lymphocyte migration: system identification antiviral cytotoxic T-cell memory in response to different stim-
approach. Immunol. Cell Biol. 2000; 78: 288–93. uli: cell population structure and protective function. Immunol.
27 Jacob J, Baltimore D. Modelling T-cell memory by genetic Cell Biol. 2001; 79: 74–86.
marking of memory T cells in vivo. Nature 1999; 399: 593–7. 49 Harada K, Ikegami T. Evolution of specificity in an immune net-
28 Bocharov GA. Modelling the dynamics of LCMV infection in work. J. Theor. Biol. 2000; 203: 439–49.
mice: conventional and exhaustive CTL responses. J. Theor. 50 Sousa J, Carneiro J. A mathematical analysis of TCR serial
Biol. 1998; 192: 283–308. triggering and down-regulation. Eur. J. Immunol. 2000; 30:
29 Edmunds WJ, van de Heijden OG, Eerola M, Gay NJ. Modelling 3219–27.
rubella in Europe. Epidemiol. Infect. 2000; 125: 617–34. 51 Brusic V, Rudy G, Honeyman G, Hammer J, Harrison L. Predic-
30 Innocent G, Morrison I, Brownlie J, Gettinby G. A computer tion of MHC class II-binding peptides using an evolutionary
simulation of the transmission dynamics and the effects of dura- algorithm and artificial neural network. Bioinformatics 1998; 14:
tion of immunity and survival of persistently infected animals on 121–30.
the spread of bovine viral diarrhoea virus in dairy cattle. Epide- 52 Lim JS, Kim S, Lee HG, Lee KY, Kwon TJ, Kim K. Selection of
miol. Infect. 1997; 119: 91–100. peptides that bind to the HLA-A2.1 molecule by molecular
31 Agur Z, Mehr R. Modelling Trypanosoma congolense parasitae- modelling. Mol. Immunol. 1996; 33: 221–30.
mia patterns during the chronic phase of infection in N’Dama 53 Honeyman MC, Brusic V, Harrison LC. Strategies for identify-
cattle. Parasite Immunol. 1997; 19: 171–82. ing and predicting islet autoantigen T-cell epitopes in insulin-
32 Gupta S, Day KP. A theoretical framework for the immuno- dependent diabetes mellitus. Ann. Med. 1997; 29: 401–4.
epidemiology of Plasmodium falciparum malaria. Parasite 54 Brusic V, Bucci K, Schonbach C, Petrovsky N, Zeleznikow J,
Immunol. 1994; 16: 361–70. Kazura JW. Efficient discovery of immune response targets by
33 Cross AP, Singer B. Modelling the development of resistance of cyclical refinement of QSAR models of peptide binding. J. Mol.
Plasmodium falciparum to anti-malarial drugs. Trans. R. Soc. Graph. Model. 2001; 19: 467.
Trop. Med. Hyg. 1991; 85: 349–55. 55 Forster MJ, Rijpkema SG. Selection of putative immunogenic
34 McLean AR, Blower SM, Modelling HIV vaccination. Trends peptides by molecular modelling of the urease of Helicobacter
Microbiol. 1995; 3: 458–62. pylori. Dev. Biol. (Basel) 2000; 103: 75–9.
35 Hraba T, Dolezal J. Mathematical modelling of HIV infection 56 Brusic V, van Endert P, Zeleznikow J, Daniel S, Hammer J,
therapy. Int. J. Immunopharmacol. 1995; 17: 523–6. Petrovsky N. A neural network model approach to the study of
36 Suppo C, Naulin JM, Langlais M, Artois M. A modelling human TAP transporter. In Silico Biol. 1999; 1: 109–21.
approach to vaccination and contraception programmes for 57 Helmer-Citterich M, Rovida E, Luzzago A, Tramontano A.
rabies control in fox populations. Proc. R. Soc. Lond. B Biol. Sci. Modelling antibody–antigen interactions: ferritin as a case study.
2000; 267: 1575–82. Mol. Immunol. 1995; 32: 1001–10.
37 Mossong J, O’Callaghan CJ, Ratnam S. Modelling antibody 58 Padlan EA, Helm BA. Modelling study of IgE/receptor inter-
response to measles vaccine and subsequent waning of immunity actions. Biochem. Soc. Trans. 1993; 21: 963–7.
in a low exposure population. Vaccine 2000; 19: 523–9. 59 Moustakas AK, Routsias J, Papadopoulos GK. Modelling of the
38 Brisson M, Edmunds WJ, Gay NJ, Law B, De Serres G. Model- MHC II allele I-A (g7) of NOD mouse: pH-dependent changes in
ling the impact of immunization on the epidemiology of varicella specificity at pockets 9 and 6 explain several of the unique prop-
zoster virus. Epidemiol. Infect. 2000; 125: 651–69. erties of this molecule. Diabetologia 2000; 43: 609–24.
 254 N Petrovsky and V Brusic

60 Tenette-Souaille C, Smith JC. Structure of the Malpha2–3 toxin 71 Baxter AG. Modelling the effects of genetic and environmental
alpha antibody-antigen complex: combination of modelling with factors on the risk of autoimmune disease. J. Autoimmun. 2001;
functional mapping experimental results. Protein Eng. 2000; 13: 16: 331–5.
345–51. 72 Nani FK, Oguztoreli MN. Modelling and simulation of
61 Irnaten M, Gallet X, Festy F et al. Prediction of epitopes and Rosenberg-type adoptive cellular immunotherapy. IMA J. Math.
production of mAb against gastric H,K-ATPase. Protein Eng. Appl. Med. Biol. 1994; 11: 107–47.
1998; 11: 949–55. 73 Owen MR, Sherratt JA. Modelling the macrophage invasion of
62 Paulino M, Esteves A, Vega M, Tabares G, Ehrlich R, Tapia O. tumours: effects on growth and composition. IMA J. Math. Appl.
Modelling a 3D structure for EgDf1 from Echinococcus granulo- Med. Biol. 1998; 15: 165–85.
sus. Putative epitopes, phosphorylation motifs and ligand. J. 74 Zarour HM, Maillere B, Brusic V et al. NY-ESO-1 119–143 is a
Comput. Aided Mol. Des. 1998; 12: 351–60. promiscuous major histocompatibility complex class II T-helper
63 Quesada M, Puig J, Delgado JM, Hidalgo-Alvarez R. Modelling epitope recognized by Th1- and Th2-type tumor-reactive CD4+
the kinetics of antigen–antibody reactions at particle enhanced T cells. Cancer Res. 2002; 62: 213–18.
optical immunoassays. J. Biomater. Sci. Polym. Ed. 1998; 9: 75 Zarour HM, Storkus WJ, Brusic V, Williams E, Kirkwood JM.
961–71. NY-ESO-1 encodes DRB1*0401-restricted epitopes recognized
64 Schiweck W, Skerra A. The rational construction of an antibody by melanoma-reactive CD4+ T cells. Cancer Res. 2000; 60:
against cystatin: lessons from the crystal structure of an artificial 4946–52.
Fab fragment. J. Mol. Biol. 1997; 268: 934–51. 76 Zarour HM, Kirkwood JM, Kierstead LS et al. Melan-A/MART-1
65 Hogeweg P. Local T–T cell and T–B cell interactions: a cellular (51–73) represents an immunogenic HLA-DR4-restricted epitope
automaton approach. Immunol. Lett. 1989; 22: 113–22. recognized by melanoma-reactive CD4 (+) T cells. Proc. Natl
66 Hodgkin PD, Chin SH, Bartell G et al. The importance of effi- Acad. Sci. USA 2000; 97: 400–5.
cacy and partial agonism in evaluating models of B lymphocyte 77 Dvorchik I, Subotin M, Marsh W, McMichael J, Fung JJ. Per-
activation. Int. Rev. Immunol. 1997; 15: 101–27. formance of multi-layer feedforward neural networks to predict
67 Genin E, Babron MC, McDermott MF et al. Modelling the major liver transplantation outcome. Meth. Inf. Med. 1996; 35: 12–18.
histocompatibility complex susceptibility to RA using the 78 Parmanto B, Doyle HR. Recurrent neural networks for predicting
MASC method. Genet. Epidemiol. 1998; 15: 419–30. outcomes after liver transplantation: representing temporal sequence
68 Brass A, Worthington J, Chen Y, Morgan K. Identification of of clinical observations. Meth. Inf. Med. 2001; 40: 386–91.
antibody epitopes within the CB-11 peptide of type II collagen. 79 Keiding N, Klein JP, Horowitz MM. Multi-state models and
II. Computer modelling studies of peptides and the interpretation outcome prediction in bone marrow transplantation. Stat. Med.
of epitope scanning results. Autoimmunity 1991; 10: 209–16. 2001; 20: 1871–85.
69 Hoover ML, Marta RT. Molecular modelling of HLA-DQ 80 Petrovsky N, Tam S, Brusic V, Russ G, Socha L, Bajic V. Use of
suggests a mechanism of resistance in type 1 diabetes. Scand. J. artificial neural networks in improving renal transplantation out-
Immunol. 1997; 45: 193–202. comes. Graft 2002; 4: 6–13.
70 Waniewski J, Prikrylova D. Autoimmunity and its therapy:
mathematical modelling. Immunol. Lett. 1988; 18: 77–80.

View publication stats