Generally speaking, MHC I binds and presents epitopes which are derived from proteolytically degraded intracellular proteins (e.g., from intracellular pathogens) and are 8–11 residues long. In contrast, MHC II epitopes are derived from extracellular sources (e.g., from extracellular pathogens), and are much longer on average (up to 25 residues [ 1 ]). Originally, it was thought that these peptide epitopes would be recognized at least in part by their secondary structure, but more recent structural data suggest that they are presented mostly in an extended form. Early prediction tools working under the wrong assumption accordingly gave inconsistent results [ 2]. Additionally, MHC I and MHC II bind peptides very differently: as the molecular structure of MHC II requires longer peptides, due to its “open” binding pocket, the residues extending the binding pocket on both sides contribute to the overall peptide binding affinity [ 1 , 3 ]. To address this finding, modern MHC II epitope prediction tools often identify a binding core, i.e., a shorter subsequence within the longer peptide sequences of the query, which is predicted to bind to the pocket.

To use prediction tools efficiently for vaccine design , one has to consider that the human MHC molecules are encoded in a highly polymorphic locus called the human leukocyte antigen (HLA) locus on chromosome 6. There are tremendous numbers of HLA alleles with different binding affinities to the same epitope sequence: more than 10,000 different human alleles have been identified and, to complicate things even further, within different populations, different alleles (i.e., variants) of the MHC genes are present in different ratios.

Various online methods are available for the prediction of epi topes, ranging from sequence-based to structure-based (using, e.g., homology modeling or docking) methods. Table 1 shows a selection of sequence-based bioinformatics tools used for MHC I or MHC II predictions, which have the advantage of speed over structure-based methods and are therefore more favorable for large-scale analysis of peptides.

Table1. Methods: QM: quantitative matrix-based methods (QM combine a matrix-based approach with a strategy to quantify the prediction scores), ANN: artificial neural networks

State-of-the-art sequence-based approaches attempt to predict the binding quality of a query sequence by abstracting from the sequence information of peptides with experimentally determined binding affinities. By doing so, they are able to generate models for each individual MHC variant. Matrix-based methods try to derive position-specific binding coefficients for each residue from a data base of known binders of the same length. For the prediction, each position of a query sequence is evaluated individually, yielding a score of congruousness to its respective position in the abstract model of a binding sequence. To predict the binding quality of the complete query sequence, the final score is given as the sum of the scores of the individual positions. This approach can be modified by adding weights to certain positions (so-called anchor positions) to increase their impact on the final score.

A second group of prediction tools relies on machine learning approaches or stochastic models like support vector machines, artificial neural networks, or hidden Markov models to predict the binding quality of a query sequence. Generally speaking, all of these approaches attempt to refi ne a model by adjusting internal parameters to the sequence information provided by a collection of known binders. Therefore, a set of known binders is used to train the model, i.e., to adjust internal parameters in such a way as to enable accurate prediction of binding quality based on empirical data (supervised learning).

Some tools in both groups also include strategies to quantitatively predict the binding of a query sequence. By incorporating either position-specific affinity contributions (matrix-based approaches) or statistical regression analysis (machine learning approaches), the user can readily compare experimentally determined IC 50 or K d values with predicted ones. However, there are no predefined absolute threshold values clearly separating query sequences into either binders or non-binders. Rather, it is advisable to defi ne cutoff values for each MHC allele individually [ 4 ].

It is important to note that all the tools, regardless of approach, heavily rely on experimental data on the measured binding affinities of peptide sequences for a specific MHC variant. Therefore, the quality of the prediction is determined by how well the binding space of a particular MHC variant is explored by the available data.

Unfortunately, for many alleles data are scarce; this has led to the development of pan-specific methods for MHC binding prediction. These use known MHC binders to known MHC alleles to infer binding for unknown pairs. Typically, such approaches are based on structural data where alleles with similar physicochemical attributes in the binding pocket are classed together using machine learning approaches [ 5 – 6 ].

References

-----------

[1] Wang P, Sidney J, Dow C et al (2008) A systematic assessment of MHC class II peptide binding predictions and evaluation of a consensus approach. PLoS Comput Biol 4, e000048

[2] Flower DR (2009) Bioinformatics for vaccinology. Wiley, Chichester, UK

[3] Zhang L, Udaka K, Mamitsuka H et al (2012) Toward more accurate pan-specific MHC- peptide binding prediction: a review of current methods and tools. Brief Bioinform 13:350–364

[4] Paul S, Weiskopf D, Angelo MA et al (2013) HLA class I alleles are associated with peptide- binding repertoires of different size, affinity, and immunogenicity. J Immunol 191: 5831–5839

[5] Doytchinova IA, Guan P, Flower DR (2004) Identifying human MHC supertypes using bioinformatic methods. J Immunol 172:4314–4323

[6] Doytchinova IA, Flower DR (2005) In silico identification of supertypes for class II MHCs. J Immunol 174:7085–7095

مواضيع ذات صلة

PROCESS OF PHAGOCYTOSIS

GRANULOCYTIC CELLS

B Cell Epitope Binding Predictions

Immune Mechanisms of Tumor Rejection

Tumor Antigens

Monoclonal antibodies

Molecular Basis of Antigen - Antibody Reactions

Role of Infections and Other Environmental Influences in Autoimmunity

Genetic Factors of Autoimmunity

الحساسية الدوائية نحو العوامل المخصرة للوصل العَصَبِي العَضَلِي

آليات حساسية السلفوناميدات ومظاهرها

Antigen - Antibody Interaction: Specificity and Cross - Reactivity