Application of an artificial neural network to predict specific class I MHC binding peptide sequences

Yeast display platform with expression of linear peptide epitopes for high-throughput assessment of peptide-MHC-II binding

Yeast display can serve as a powerful tool to assess the binding of peptides to the major histocompatibility complex (pMHC) and pMHC-T-cell receptor binding. However, this approach is often limited by the need to optimize MHC proteins for yeast surface expression, which can be laborious and may not yield productive results. Here we present a second-generation yeast display platform for class II MHC molecules (MHC-II), which decouples MHC-II expression from yeast-expressed peptides, referred to as "peptide display." Peptide display obviates the need for yeast-specific MHC optimizations and increases the scale of MHC-II alleles available for use in yeast display screens. Because MHC identity is separated from the peptide library, a further benefit of this platform is the ability to assess a single library of peptides against any MHC-II. We demonstrate the utility of the peptide display platform across MHC-II proteins, screening HLA-DR, HLA-DP, and HLA-DQ alleles. We further explore parameters of selections, including reagent dependencies, MHC avidity, and use of competitor peptides. In summary, this approach presents an advance in the throughput and accessibility of screening peptide-MHC-II binding.

Immunoinformatics: Predicting Peptide-MHC Binding

Immunoinformatics is a discipline that applies methods of computer science to study and model the immune system. A fundamental question addressed by immunoinformatics is how to understand the rules of antigen presentation by MHC molecules to T cells, a process that is central to adaptive immune responses to infections and cancer. In the modern era of personalized medicine, the ability to model and predict which antigens can be presented by MHC is key to manipulating the immune system and designing strategies for therapeutic intervention. Since the MHC is both polygenic and extremely polymorphic, each individual possesses a personalized set of MHC molecules with different peptide-binding specificities, and collectively they present a unique individualized peptide imprint of the ongoing protein metabolism. Mapping all MHC allotypes is an enormous undertaking that cannot be achieved without a strong bioinformatics component. Computational tools for the prediction of peptide-MHC binding have thus become essential in most pipelines for T cell epitope discovery and an inescapable component of vaccine and cancer research. Here, we describe the development of several such tools, from pioneering efforts to the current state-of-the-art methods, that have allowed for accurate predictions of peptide binding of all MHC molecules, even including those that have not yet been characterized experimentally.

T Cell Epitope Predictions

Throughout the body, T cells monitor MHC-bound ligands expressed on the surface of essentially all cell types. MHC ligands that trigger a T cell immune response are referred to as T cell epitopes. Identifying such epitopes enables tracking, phenotyping, and stimulating T cells involved in immune responses in infectious disease, allergy, autoimmunity, transplantation, and cancer. The specific T cell epitopes recognized in an individual are determined by genetic factors such as the MHC molecules the individual expresses, in parallel to the individual's environmental exposure history. The complexity and importance of T cell epitope mapping have motivated the development of computational approaches that predict what T cell epitopes are likely to be recognized in a given individual or in a broader population. Such predictions guide experimental epitope mapping studies and enable computational analysis of the immunogenic potential of a given protein sequence region.

Development, Validation and Comparison of Artificial Neural Network Models and Logistic Regression Models Predicting Survival of Unresectable Pancreatic Cancer

Background: Prediction models for the overall survival of pancreatic cancer remain unsatisfactory. We aimed to explore artificial neural networks (ANNs) modeling to predict the survival of unresectable pancreatic cancer patients.

Methods: Thirty-two clinical parameters were collected from 221 unresectable pancreatic cancer patients, and their prognostic ability was evaluated using univariate and multivariate logistic regression. ANN and logistic regression (LR) models were developed on a training group (168 patients), and the area under the ROC curve (AUC) was used for comparison of the ANN and LR models. The models were further tested on the testing group (53 patients), and k-statistics were used for accuracy comparison.

Results: We built three ANN models, based on 3, 7, and 32 basic features, to predict 8 month survival. All 3 ANN models showed better performance, with AUCs significantly higher than those from the respective LR models (0.811 vs. 0.680, 0.844 vs. 0.722, 0.921 vs. 0.849, all p < 0.05). The ability of the ANN models to discriminate 8 month survival with higher accuracy than the respective LR models was further confirmed in 53 consecutive patients.

Conclusion: We developed ANN models predicting the 8 month survival of unresectable pancreatic cancer patients. These models may help to optimize personalized patient management.

Fundamentals and Methods for T- and B-Cell Epitope Prediction

Adaptive immunity is mediated by T- and B-cells, which are immune cells capable of developing pathogen-specific memory that confers immunological protection. Memory and effector functions of B- and T-cells are predicated on the recognition through specialized receptors of specific targets (antigens) in pathogens. More specifically, B- and T-cells recognize portions within their cognate antigens known as epitopes. There is great interest in identifying epitopes in antigens for a number of practical reasons, including understanding disease etiology, immune monitoring, developing diagnosis assays, and designing epitope-based vaccines. Epitope identification is costly and time-consuming as it requires experimental screening of large arrays of potential epitope candidates. Fortunately, researchers have developed in silico prediction methods that dramatically reduce the burden associated with epitope mapping by decreasing the list of potential epitope candidates for experimental testing. Here, we analyze aspects of antigen recognition by T- and B-cells that are relevant for epitope prediction. Subsequently, we provide a systematic and inclusive review of the most relevant B- and T-cell epitope prediction methods and tools, paying particular attention to their foundations.

Scrutinizing MHC-I binding peptides and their limits of variation

Designed peptides that bind to major histocompatibility protein I (MHC-I) allomorphs bear the promise of representing epitopes that stimulate a desired immune response. A rigorous bioinformatical exploration of sequence patterns hidden in peptides that bind to the mouse MHC-I allomorph H-2K^b is presented. We exemplify and validate these motif findings by systematically dissecting the epitope SIINFEKL and analyzing the resulting fragments for their binding potential to H-2K^b in a thermal denaturation assay. The results demonstrate that only fragments exclusively retaining the carboxy- or amino-terminus of the reference peptide exhibit significant binding potential, with the N-terminal pentapeptide SIINF as shortest ligand. This study demonstrates that sophisticated machine-learning algorithms excel at extracting fine-grained patterns from peptide sequence data and predicting MHC-I binding peptides, thereby considerably extending existing linear prediction models and providing a fresh view on the computer-based molecular design of future synthetic vaccines. The server for prediction is available at http://modlab-cadd.ethz.ch (SLiDER tool, MHC-I version 2012).

Future success in vaccine development will critically depend on identifying potent epitopes with reduced side effects. Among such candidate molecules, immunogenic peptides binding to major histocompatibility protein I (MHC-I) represent a preferred class of biomolecules for vaccine design. Computational models assist in the selection of the best candidate peptides by providing a mathematical rationale for antigen recognition by MHC-I. Here we present a machine-learning model that was trained on recognizing features of known MHC-I binding and non-binding peptide sequences with sustained accuracy. We were able to biochemically validate the computational predictions in a direct binding assay measuring complex formation between synthesized candidate peptides and MHC-I. Strong correspondence between the predictions and the experimentally determined binding potential corroborate the machine-learning model as viable for future antigen design. Thus, our study provides a concept for rapidly finding innovative MHC-I binding peptides with limited experimental effort.

Computational Resources for MHC Ligand Identification

Advances in the high-throughput determination of functional modulators of major histocompatibility complex (MHC) and improved computational predictions of MHC ligands have rendered the rational design of immunomodulatory peptides feasible. Proteome-derived peptides and 'reverse vaccinology' by computational means will play a driving role in future vaccine design. Here we review the molecular mechanisms of the MHC mediated immune response, present the computational approaches that have emerged in this area of biotechnology, and provide an overview of publicly available computational resources for predicting and designing new peptidic MHC ligands.

Molecular evolution of a peptide GPCR ligand driven by artificial neural networks

Peptide ligands of G protein-coupled receptors constitute valuable natural lead structures for the development of highly selective drugs and high-affinity tools to probe ligand-receptor interaction. Currently, pharmacological and metabolic modification of natural peptides involves either an iterative trial-and-error process based on structure-activity relationships or screening of peptide libraries that contain many structural variants of the native molecule. Here, we present a novel neural network architecture for the improvement of metabolic stability without loss of bioactivity. In this approach the peptide sequence determines the topology of the neural network and each cell corresponds one-to-one to a single amino acid of the peptide chain. Using a training set, the learning algorithm calculated weights for each cell. The resulting network calculated the fitness function in a genetic algorithm to explore the virtual space of all possible peptides. The network training was based on gradient descent techniques which rely on the efficient calculation of the gradient by back-propagation. After three consecutive cycles of sequence design by the neural network, peptide synthesis and bioassay this new approach yielded a ligand with 70fold higher metabolic stability compared to the wild type peptide without loss of the subnanomolar activity in the biological assay. Combining specialized neural networks with an exploration of the combinatorial amino acid sequence space by genetic algorithms represents a novel rational strategy for peptide design and optimization.

Peptide binding motifs for MHC class I and II molecules

This overview discusses the use of peptide-binding motifs to predict interaction with a specific MHC class I or II allele, and gives examples for the use of MHC binding motifs to predict T-cell recognition.

Robust quantitative modeling of peptide binding affinities for MHC molecules using physical-chemical descriptors

Major histocompatibility complex (MHC) molecules bind short peptides resulting from intracellular processing of foreign and self proteins, and present them on the cell surface for recognition by T-cell receptors. We propose a new robust approach to quantitatively model the binding affinities of MHC molecules by quantitative structure-activity relationships (QSAR) that use the physical-chemical amino acid descriptors E1-E5. These QSAR models are robust, sequence-based, and can be used as a fast and reliable filter to predict the MHC binding affinity for large protein databases.

A genetic approach for building different alphabets for peptide and protein classification

Background

In this paper, it is proposed an optimization approach for producing reduced alphabets for peptide classification, using a Genetic Algorithm. The classification task is performed by a multi-classifier system where each classifier (Linear or Radial Basis function Support Vector Machines) is trained using features extracted by different reduced alphabets. Each alphabet is constructed by a Genetic Algorithm whose objective function is the maximization of the area under the ROC-curve obtained in several classification problems.

Results

The new approach has been tested in three peptide classification problems: HIV-protease, recognition of T-cell epitopes and prediction of peptides that bind human leukocyte antigens. The tests demonstrate that the idea of training a pool classifiers by reduced alphabets, created using a Genetic Algorithm, allows an improvement over other state-of-the-art feature extraction methods.

Conclusion

The validity of the novel strategy for creating reduced alphabets is demonstrated by the performance improvement obtained by the proposed approach with respect to other reduced alphabets-based methods in the tested problems.

Leveraging information across HLA alleles/supertypes improves epitope prediction

We present a model for predicting HLA class I restricted CTL epitopes. In contrast to almost all other work in this area, we train a single model on epitopes from all HLA alleles and supertypes, yet retain the ability to make epitope predictions for specific HLA alleles. We are therefore able to leverage data across all HLA alleles and/or their supertypes, automatically learning what information should be shared and also how to combine allele-specific, supertype-specific, and global information in a principled way. We show that this leveraging can improve prediction of epitopes having HLA alleles with known supertypes, and dramatically increases our ability to predict epitopes having alleles which do not fall into any of the known supertypes. Our model, which is based on logistic regression, is simple to implement and understand, is solved by finding a single global maximum, and is more accurate (to our knowledge) than any other model.

Structure-based prediction of MHC-peptide association: algorithm comparison and application to cancer vaccine design

Peptide vaccination for cancer immunotherapy requires identification of peptide epitopes derived from antigenic proteins associated with the tumor. Such peptides can bind to MHC proteins (MHC molecules) on the tumor-cell surface, with the potential to initiate a host immune response against the tumor. Computer prediction of peptide epitopes can be based on known motifs for peptide sequences that bind to a certain MHC molecule, on algorithms using experimental data as a training set, or on structure-based approaches. We have developed an algorithm, which we refer to as PePSSI, for flexible structural prediction of peptide binding to MHC molecules. Here, we have applied this algorithm to identify peptide epitopes (of nine amino acids, the common length) from the sequence of the cancer-testis antigen KU-CT-1, based on the potential of these peptides to bind to the human MHC molecule HLA-A2. We compared the PePSSI predictions with those of other algorithms and found that several peptides predicted to be strong HLA-A2 binders by PePSSI were similarly predicted by another structure-based algorithm, PREDEP. The results show how structure-based prediction can identify potential peptide epitopes without known binding motifs and suggest that side chain orientation in binding peptides may be obtained using PePSSI.

Immunoinformatics comes of age

With the burgeoning immunological data in the scientific literature, scientists must increasingly rely on Internet resources to inform and enhance their work. Here we provide a brief overview of the adaptive immune response and summaries of immunoinformatics resources, emphasizing those with Web interfaces. These resources include searchable databases of epitopes and immune-related molecules, and analysis tools for T cell and B cell epitope prediction, vaccine design, and protein structure comparisons. There is an agreeable synergy between the growing collections in immune-related databases and the growing sophistication of analysis software; the databases provide the foundation for developing predictive computational tools, which in turn enable more rapid identification of immune responses to populate the databases. Collectively, these resources contribute to improved understanding of immune responses and escape, and evolution of pathogens under immune pressure. The public health implications are vast, including designing vaccines, understanding autoimmune diseases, and defining the correlates of immune protection.

Toward a multiscale model of antigen presentation in immunity

A functioning immune system and the process of antigen presentation in particular encompass events that occur at multiple length and time scales. Despite a wealth of information in the biological literature regarding each of these scales, no single representation synthesizing this information into a model of the overall immune response as it depends on antigen presentation is available. In this article, we outline an approach for integrating information over relevant biological and temporal scales to generate such a representation for major histocompatibility complex class II-mediated antigen presentation. In addition, we begin to address how such models can be used to answer questions about mechanisms of infection and new strategies for treatment and vaccines.

Immunogenicity screening in protein drug development

Most therapeutic proteins in clinical trials or on the market are, to a variable extent, immunogenic. Formation of antidrug antibodies poses a risk that should be assessed during drug development, as it possibly compromises drug safety and alters pharmacokinetics. The generation of these antibodies is critically dependent on the presence of T helper cell epitopes, which have classically been identified by in vitro methods using blood cells from human donors. As a novel development, in silico methods that identify T cell epitopes have been coming on line. These methods are relatively inexpensive and allow the mapping of epitopes from virtually all human leukocyte antigen molecules derived from a wide genetic background. In vitro assays remain important, but guided by in silico data they can focus on selected peptides and human leukocyte antigen haplotypes, thereby significantly reducing time and cost.

HLA-peptide binding prediction using structural and modeling principles

Short peptides binding to specific human leukocyte antigen (HLA) alleles elicit immune response. These candidate peptides have potential utility in peptide vaccine design and development. The binding of peptides to allele-specific HLA molecule is estimated using competitive binding assay and biochemical binding constants. Application of this method for proteome-wide screening in parasites, viruses, and virulent bacterial strains is laborious and expensive. However, short listing of candidate peptides using prediction approaches have been realized lately. Prediction of peptide binding to HLA alleles using structural and modeling principles has gained momentum in recent years. Here, we discuss the current status of such prediction.

Ab initio prediction of peptide-MHC binding geometry for diverse class I MHC allotypes

Since determining the crystallographic structure of all peptide-MHC complexes is infeasible, an accurate prediction of the conformation is a critical computational problem. These models can be useful for determining binding energetics, predicting the structures of specific ternary complexes with T-cell receptors, and designing new molecules interacting with these complexes. The main difficulties are (1) adequate sampling of the large number of conformational degrees of freedom for the flexible peptide, (2) predicting subtle changes in the MHC interface geometry upon binding, and (3) building models for numerous MHC allotypes without known structures. Whereas previous studies have approached the sampling problem by dividing the conformational variables into different sets and predicting them separately, we have refined the Biased-Probability Monte Carlo docking protocol in internal coordinates to optimize a physical energy function for all peptide variables simultaneously. We also imitated the induced fit by docking into a more permissive smooth grid representation of the MHC followed by refinement and reranking using an all-atom MHC model. Our method was tested by a comparison of the results of cross-docking 14 peptides into HLA-A*0201 and 9 peptides into H-2K(b) as well as docking peptides into homology models for five different HLA allotypes with a comprehensive set of experimental structures. The surprisingly accurate prediction (0.75 A backbone RMSD) for cross-docking of a highly flexible decapeptide, dissimilar to the original bound peptide, as well as docking predictions using homology models for two allotypes with low average backbone RMSDs of less than 1.0 A illustrate the method's effectiveness. Finally, energy terms calculated using the predicted structures were combined with supervised learning on a large data set to classify peptides as either HLA-A*0201 binders or nonbinders. In contrast with sequence-based prediction methods, this model was also able to predict the binding affinity for peptides to a different MHC allotype (H-2K(b)), not used for training, with comparable prediction accuracy.

Design of enhanced agonists through the use of a new virtual screening method: application to peptides that bind class I major histocompatibility complex (MHC) molecules

A new screening procedure is described that uses docking calculations to design enhanced agonist peptides that bind to major histocompatibility complex (MHC) class I receptors. The screening process proceeds via single mutations of one amino acid at the positions that directly interact with the MHC receptor. The energetic and structural effects of these mutations have been studied using fragments of the original ligand that vary in length. The results of these docking studies indicate that the mutant affinity ranking of long peptides can be practically reproduced with a screening approach performed using fragments of six residues. Fragments of four and five residues could mimic, in some cases, the structural arrangement of the side chains of the full-length peptide. We have compared the structural and energetic results of the docking calculations with experimental data using three unrelated ligand peptides that differ greatly in their affinity for the MHC complex. Analysis of the affinity of the fragments led to the identification of three important parameters in the construction of fragments that mimic the structural and energetic properties of the full-length ligand: the length of the fragment; its intermolecular energy; and the number and localization, internal or terminal, of the anchor residues. The results of this new peptide-design methodology have been applied to suggest new peptides derived from the MUC1-8 peptide that could be used as murine vaccines that trigger the immune response through the MHC class I protein H-2K(b).

Computational methods for prediction of T-cell epitopes--a framework for modelling, testing, and applications

Computational models complement laboratory experimentation for efficient identification of MHC-binding peptides and T-cell epitopes. Methods for prediction of MHC-binding peptides include binding motifs, quantitative matrices, artificial neural networks, hidden Markov models, and molecular modelling. Models derived by these methods have been successfully used for prediction of T-cell epitopes in cancer, autoimmunity, infectious disease, and allergy. For maximum benefit, the use of computer models must be treated as experiments analogous to standard laboratory procedures and performed according to strict standards. This requires careful selection of data for model building, and adequate testing and validation. A range of web-based databases and MHC-binding prediction programs are available. Although some available prediction programs for particular MHC alleles have reasonable accuracy, there is no guarantee that all models produce good quality predictions. In this article, we present and discuss a framework for modelling, testing, and applications of computational methods used in predictions of T-cell epitopes.

Sensitive quantitative predictions of peptide-MHC binding by a 'Query by Committee' artificial neural network approach

We have generated Artificial Neural Networks (ANN) capable of performing sensitive, quantitative predictions of peptide binding to the MHC class I molecule, HLA-A*0204. We have shown that such quantitative ANN are superior to conventional classification ANN, that have been trained to predict binding vs non-binding peptides. Furthermore, quantitative ANN allowed a straightforward application of a 'Query by Committee' (QBC) principle whereby particularly information-rich peptides could be identified and subsequently tested experimentally. Iterative training based on QBC-selected peptides considerably increased the sensitivity without compromising the efficiency of the prediction. This suggests a general, rational and unbiased approach to the development of high quality predictions of epitopes restricted to this and other HLA molecules. Due to their quantitative nature, such predictions will cover a wide range of MHC-binding affinities of immunological interest, and they can be readily integrated with predictions of other events involved in generating immunogenic epitopes. These predictions have the capacity to perform rapid proteome-wide searches for epitopes. Finally, it is an example of an iterative feedback loop whereby advanced, computational bioinformatics optimize experimental strategy, and vice versa.

Simultaneous prediction of binding capacity for multiple molecules of the HLA B44 supertype

We selected for study a set of B44-supertype molecules collectively represented in >40% of the individuals in all major ethnicities (B*1801, B*4001, B*4002, B*4402, B*4403, and B*4501). The peptide-binding specificity of each molecule was characterized using single amino acid substitution analogues and nonredundant peptide libraries. In all cases, only peptide ligands with glutamic acid in position 2 were preferred. At the C terminus, each allele was associated with a unique but broad pattern of preferences, but all molecules tolerated hydrophobic/aliphatic (leucine, isoleucine, valine, methionine), aromatic (tyrosine, phenylalanine, tryptophan), and small (alanine, glycine, threonine) residues. Secondary anchor motifs were also defined for all molecules. Together, these features were used to define a B44 supermotif and a novel algorithm for calculating degeneracy scores that can be used to predict B44-supertype degenerate binders. Approximately 90% of the peptides with a B44 supermotif degeneracy score of >10 bound at least three of the six B44-supertype molecules studied with high affinity. Finally, a number of peptides derived from hepatitis B and C viruses, HIV, and Plasmodium falciparum have been identified that have degenerate B44 supertype-binding capacity. Taken together, these findings have important implications for epitope-based approaches to vaccination, immunotherapy, and the monitoring of immune responses.

Genomewide Conserved Epitope Profiles of HIV-1 Predicted by Biophysical Properties of MHC Binding Peptides

We propose a new method for predicting MHC binding of peptides using biophysical parameters of the constituent amino acids. Unlike conventional matrix-based methods, our method does not assume independent binding of the individual side chains and uses a model that simultaneously represents all the residues. The model discovers the quantified 9-mer "property model" within the longer peptides that are most common among binders. Prediction for a new peptide is based on its statistical "distance" from the extracted peptide property model. MHC-specific peptide property models were constructed from compiled binder/nonbinder data using this method. We report the results of cross-validation of the prediction method and comparison with other methods. The comparison suggests that our method performs substantially better for some MHC class II molecules and equally well for other MHC types. To demonstrate large-scale utility, 30 HIV-1 reference genomes covering diverse subtypes were analyzed. Regions that are likely to bind MHC (A2, DR1, or DR4) and that are conserved across the HIV-1 subtypes were identified. These "epitope profiles" of the diverse HIV-1 strains can also be visually presented to facilitate discovery of conserved patterns naturally occurring in the viral genomes. As an essential step in designing vaccines, the revealed patterns may provide valuable information in identifying the immunologically important regions.

Optimization and immune recognition of multiple novel conserved HLA-A2, human immunodeficiency virus type 1-specific CTL epitopes

MHC-I-restricted cytotoxic responses are considered a critical component of protective immunity against viruses, including human immunodeficiency virus type 1 (HIV-1). CTLs directed against accessory and early regulatory HIV-1 proteins might be particularly effective; however, CTL epitopes in these proteins are rarely found. Novel artificial neural networks (ANNs) were used to quantitatively predict HLA-A2-binding CTL epitope peptides from publicly available full-length HIV-1 protein sequences. Epitopes were selected based on their novelty, predicted HLA-A2-binding affinity and conservation among HIV-1 strains. HLA-A2 binding was validated experimentally and binders were tested for their ability to induce CTL and IFN-gamma responses. About 69 % were immunogenic in HLA-A2 transgenic mice and 61 % were recognized by CD8(+) T-cells from 17 HLA-A2 HIV-1-positive patients. Thus, 31 novel conserved CTL epitopes were identified in eight HIV-1 proteins, including the first HLA-A2 minimal epitopes ever reported in the accessory and regulatory proteins Vif, Vpu and Rev. Interestingly, intermediate-binding peptides of low or no immunogenicity (i.e. subdominant epitopes) were found to be antigenic and more conserved. Such epitope peptides were anchor-optimized to improve immunogenicity and further increase the number of potential vaccine epitopes. About 67 % of anchor-optimized vaccine epitopes induced immune responses against the corresponding non-immunogenic naturally occurring epitopes. This study demonstrates the potency of ANNs for identifying putative virus CTL epitopes, and the new HIV-1 CTL epitopes identified should have significant implications for HIV-1 vaccine development. As a novel vaccine approach, it is proposed to increase the coverage of HIV variants by including multiple anchor-optimized variants of the more conserved subdominant epitopes.

Prognostic models in patients with non-small-cell lung cancer using artificial neural networks in comparison with logistic regression

It is difficult to precisely predict the outcome of each individual patient with non-small-cell lung cancer (NSCLC) by using conventional statistical methods and ordinary clinico-pathological variables. We applied artificial neural networks (ANN) for this purpose. We constructed a prognostic model for 125 NSCLC patients with 17 potential input variables, including 12 clinico-pathological variables (age, sex, smoking index, tumor size, p factor, pT, pN, stage, histology) and 5 immunohistochemical variables (p27 percentage, p27 intensity, p53, cyclin D1, retinoblastoma (RB)), by using the parameter-increasing method (PIM). Using the resultant ANN model, prediction was possible in 104 of 125 patients (83%, judgment ratio (JR)) and accuracy for prediction of survival at 5 years was 87%. On the other hand, JR and survival prediction accuracy in the logistic regression (LR) model were 37% and 78%, respectively. In addition, ANN outperformed LR for prediction of survival at 1 or 3 years. In these cases, PIM selected p27 intensity and cyclin D1 for the 3-year survival model and p53 for the 1-year survival model in addition to clinico-pathological variables. Finally, even in an independent validation data set of 48 patients, who underwent surgery 10 years later, the present ANN model could predict outcome of patients at 5 years with the JR and accuracy of 81% and 77%, respectively. This study demonstrates that ANN is a potentially more useful tool than conventional statistical methods for predicting survival of patients with NSCLC and that inclusion of relevant molecular markers as input variables enhances its predictive ability.

T-cell epitope prediction with combinatorial peptide libraries

T cell receptors (TCR) recognize antigenic peptides in complex with the major histocompatibility complex (MHC) molecules and this trimolecular interaction initiates antigen-specific signaling pathways in the responding T lymphocytes. For the study of autoimmune diseases and vaccine development, it is important to identify peptides (epitopes) that can stimulate a given TCR. The use of combinatorial peptide libraries has recently been introduced as a powerful tool for this purpose. A combinatorial library of n-mer peptides is a set of complex mixtures each characterized by one position fixed to be a specified amino acid and all other positions randomized. A given TCR can be fingerprinted by screening a variety of combinatorial libraries using a proliferation assay. Here, we present statistical models for elucidating the recognition profile of a TCR using combinatorial library proliferation assay data and known MHC binding data.

Recovery of known T-cell epitopes by computational scanning of a viral genome

A new computational method (EpiDock) is proposed for predicting peptide binding to class I MHC proteins, from the amino acid sequence of any protein of immunological interest. Starting from the primary structure of the target protein, individual three-dimensional structures of all possible MHC-peptide (8-, 9- and 10-mers) complexes are obtained by homology modelling. A free energy scoring function (Fresno) is then used to predict the absolute binding free energy of all possible peptides to the class I MHC restriction protein. Assuming that immunodominant epitopes are usually found among the top MHC binders, the method can thus be applied to predict the location of immunogenic peptides on the sequence of the protein target. When applied to the prediction of HLA-A*0201-restricted T-cell epitopes from the Hepatitis B virus, EpiDock was able to recover 92% of known high affinity binders and 80% of known epitopes within a filtered subset of all possible nonapeptides corresponding to about one tenth of the full theoretical list. The proposed method is fully automated and fast enough to scan a viral genome in less than an hour on a parallel computing architecture. As it requires very few starting experimental data, EpiDock can be used: (i) to predict potential T-cell epitopes from viral genomes (ii) to roughly predict still unknown peptide binding motifs for novel class I MHC alleles.

Combining computer algorithms with experimental approaches permits the rapid and accurate identification of T cell epitopes from defined antigens

The identification of T cell epitopes from immunologically relevant antigens remains a critical step in the development of vaccines and methods for monitoring of T cell responses. This review presents an overview of strategies that employ computer algorithms for the selection of candidate peptides from defined proteins and subsequent verification of their in vivo relevance by experimental approaches. Several computer algorithms are currently being used for epitope prediction of various major histocompatibility complex (MHC) class I and II molecules, based either on the analysis of natural MHC ligands or on the binding properties of synthetic peptides. Moreover, the analysis of proteasomal digests of peptides and whole proteins has led to the development of algorithms for the prediction of proteasomal cleavages. In order to verify the generation of the predicted peptides during antigen processing in vivo as well as their immunogenic potential, several experimental approaches have been pursued in the recent past. Mass spectrometry-based bioanalytical approaches have been used specifically to detect predicted peptides among isolated natural ligands. Other strategies employ various methods for the stimulation of primary T cell responses against the predicted peptides and subsequent testing of the recognition pattern towards target cells that express the antigen.

Majority of peptides binding HLA-A*0201 with high affinity crossreact with other A2-supertype molecules

The A*0201, A *0202, A*0203, A*0206, and A*6802 binding capacity of single amino acid substitution analogs of known A2-supertype binding peptides and of large nonredundant peptide libraries was measured. The results were utilized to rigorously define the peptide binding specificities of these A2-supertype molecules. Although each molecule was noted to have unique preferences, large overlaps in specificity were found. The presence of L, I, V, M, A, T, and Q residues in position 2, and L, I, V, M, A, and T residues at the C-terminus of peptide ligands were tolerated by all molecules. Likewise, whereas examination of secondary influences on peptide binding revealed allele specific preferences, shared features could also be identified. These shared features were utilized to define an A2-supermotif and were noted to correlate with crossreactivity. Over 70% of the peptides that bound A *0201 with high affinity were found to bind at least two other A2-supertype molecules. Because the A2-supertype molecules studied herein cover the variants most common in different major ethnicities, these findings have important implications for epitope-based approaches to vaccination, immunotherapy, and the monitoring of immune responses.

Combinatorial peptide libraries and biometric score matrices permit the quantitative analysis of specific and degenerate interactions between clonotypic TCR and MHC peptide ligands

The interaction of TCRs with MHC peptide ligands can be highly flexible, so that many different peptides are recognized by the same TCR in the context of a single restriction element. We provide a quantitative description of such interactions, which allows the identification of T cell epitopes and molecular mimics. The response of T cell clones to positional scanning synthetic combinatorial libraries is analyzed with a mathematical approach that is based on a model of independent contribution of individual amino acids to peptide Ag recognition. This biometric analysis compares the information derived from these libraries composed of trillions of decapeptides with all the millions of decapeptides contained in a protein database to rank and predict the most stimulatory peptides for a given T cell clone. We demonstrate the predictive power of the novel strategy and show that, together with gene expression profiling by cDNA microarrays, it leads to the identification of novel candidate autoantigens in the inflammatory autoimmune disease, multiple sclerosis.

Predicting sequences and structures of MHC-binding peptides: a computational combinatorial approach

Peptides bound to MHC molecules on the surface of cells convey critical information about the cellular milieu to immune system T cells. Predicting which peptides can bind an MHC molecule, and understanding their modes of binding, are important in order to design better diagnostic and therapeutic agents for infectious and autoimmune diseases. Due to the difficulty of obtaining sufficient experimental binding data for each human MHC molecule, computational modeling of MHC peptide-binding properties is necessary. This paper describes a computational combinatorial design approach to the prediction of peptides that bind an MHC molecule of known X-ray crystallographic or NMR-determined structure. The procedure uses chemical fragments as models for amino acid residues and produces a set of sequences for peptides predicted to bind in the MHC peptide-binding groove. The probabilities for specific amino acids occurring at each position of the peptide are calculated based on these sequences, and these probabilities show a good agreement with amino acid distributions derived from a MHC-binding peptide database. The method also enables prediction of the three-dimensional structure of MHC-peptide complexes. Docking, linking, and optimization procedures were performed with the XPLOR program [1].

Relating amino acid sequence to phenotype: analysis of peptide-binding data

We illustrate data analytic concerns that arise in the context of relating genotype, as represented by amino acid sequence, to phenotypes (outcomes). The present application examines whether peptides that bind to a particular major histocompatibility complex (MHC) class I molecule have characteristic amino acid sequences. However, the concerns identified and addressed are considerably more general. It is recognized that simple rules for predicting binding based solely on preferences for specific amino acids in certain (anchor) positions of the peptide's amino acid sequence are generally inadequate and that binding is potentially influenced by all sequence positions as well as between-position interactions. The desire to elucidate these more complex prediction rules has spawned various modeling attempts, the shortcomings of which provide motivation for the methods adopted here. Because of (i) this need to model between-position interactions, (ii) amino acids constituting a highly (20) multilevel unordered categorical covariate, and (iii) there frequently being numerous such covariates (i.e., positions) comprising the sequence, standard regression/classification techniques are problematic due to the proliferation of indicator variables required for encoding the sequence position covariates and attendant interactions. These difficulties have led to analyses based on (continuous) properties (e.g., molecular weights) of the amino acids. However, there is potential information loss in such an approach if the properties used are incomplete and/or do not capture the mechanism underlying association with the phenotype. Here we demonstrate that handling unordered categorical covariates with numerous levels and accompanying interactions can be done effectively using classification trees and recently devised bump-hunting methods. We further tackle the question of whether observed associations are attributable to amino acid properties as well as addressing the assessment and implications of between-position covariation.

Towards the MHC-peptide combinatorics

The exponentially increased sequence information on major histocompatibility complex (MHC) alleles points to the existence of a high degree of polymorphism within them. To understand the functional consequences of MHC alleles, 36 nonredundant MHC-peptide complexes in the protein data bank (PDB) were examined. Induced fit molecular recognition patterns such as those in MHC-peptide complexes are governed by numerous rules. The 36 complexes were clustered into 19 subgroups based on allele specificity and peptide length. The subgroups were further analyzed for identifying common features in MHC-peptide binding pattern. The four major observations made during the investigation were: (1) the positional preference of peptide residues defined by percentage burial upon complex formation is shown for all the 19 subgroups and the burial profiles within entries in a given subgroup are found to be similar; (2) in class I specific 8- and 9-mer peptides, the fourth residue is consistently solvent exposed, however this observation is not consistent in class I specific 10-mer peptides; (3) an anchor-shift in positional preference is observed towards the C terminal as the peptide length increases in class II specific peptides; and (4) peptide backbone atoms are proportionately dominant at the MHC-peptide interface.

Definition of the Mamu A*01 peptide binding specificity: application to the identification of wild-type and optimized ligands from simian immunodeficiency virus regulatory proteins

Single amino acid substitution analogs of the known Mamu A*01 binding peptide gag 181-190 and libraries of naturally occurring sequences of viral or bacterial origin were used to rigorously define the peptide binding motif associated with Mamu A*01 molecules. The presence of S or T in position 2, P in position 3, and hydrophobic or aromatic residues at the C terminus is associated with optimal binding capacity. At each of these positions, additional residues are also tolerated but associated with significant decreases in binding capacity. The presence of at least two preferred and one tolerated residues at the three anchor positions is necessary for good Mamu A*01 binding; optimal ligand size is 8-9 residues. This detailed motif has been used to map potential epitopes from SIVmac239 regulatory proteins and to engineer peptides with increased binding capacity. A total of 13 wild type and 17 analog candidate epitopes were identified. Furthermore, our analysis reveals a significantly lower than expected frequency of epitopes in early regulatory proteins, suggesting a possible evolutionary- and/or immunoselection directed against variants of viral products that contain CTL epitopes.

Description and prediction of peptide-MHC binding: the 'human MHC project'

MHC molecules are crucially involved in controlling the specific immune system. They are highly polymorphic receptors sampling peptides from the cellular environment and presenting these peptides for scrutiny by immune cells. Recent advances in combinatorial peptide chemistry have improved the description and prediction of peptide-MHC binding. It is envisioned that a complete mapping of human immune reactivities will be possible.