Peptide-MHC Binding Prediction

Many researchers try to find personalized treatment for cancers through major histocompatibility complex (MHC, also called Human Leukocyte Antigen (HLA) complex), which is capable of binding peptides derived from intracellular proteins and displaying them at the cell surface. Neoantigens are ideal targets for immunotherapy. It's necessary to understand the binding affinity between specific peptides and MHC alleles. During the protein drug therapy and vaccine development, immunogenicity is a big obstacle to the success. A basic approach for immunogenicity prediction is based on the premise that effective neoepitope, which is a kind of unique peptide signatures expressed on infected or tumoral cells and recognized by T cells, should bind with the MHC (class I MHC molecule) with high affinity. It's remaining a challenging step in cancer therapy to select immunogenic neoepitopes. With the improvement of sequencing technology, bioinformatics and artificial intelligence, deep learning models developed by researchers are valuable for screening effective neoepitopes in cancer vaccine development.

Prerequisites for Peptide-MHC Binding Affinity Prediction with High Accuracy

Availability of large-scale training datasets
Application of artificial neural networks, especially Recurrent Neural Networks (RNNs)

Figure 1 An overview of the MHCSeqNet’s architecture. (Poomarin Phloyphisut, et al. 2019)

Software Tools for Peptide-MHC Binding Affinity Prediction

State-of-the-art software tools employed deep learning models (NetMHCpan, ConvMHC, MHCflurry, MHCSeqNet)

Software Tools	Input	Method
NetMHCpan	▪ variable-length inputs of any peptide ▪ MHC alleles with known AA sequence	▪ artificial neural networks (ANNs) ▪ position-specific encoding system based on amino acid substitution matrices
ConvMHC	▪ fixed-length inputs of peptide epitopes ▪ specific MHC alleles	deep convolutional neural network (DCNN)
MHCflurry	▪ fixed-length inputs of peptide epitopes ▪ specific MHC alleles	▪ ensembles of allele-specific models ▪ position-specific encoding system based on amino acid substitution matrices
MHCSeqNet	▪ no restriction on the input peptide or MHC allele with known AA sequence	▪ deep learning model, recurrent neural networks (RNNs) ▪ context-aware amino acid embedding model

* Requirements of inputs: peptide, in the form of amino acid sequence, and MHC allele, in the form of either amino acid sequence or allele name

Table 1 Software tools used in MHC binding prediction

Training and Testing Datasets

Immune Epitope Database (IEDB)
MHC ligand peptidome datasets
Large amino acid database
Natural ligand datasets from mass spectrometry data
Recent publications

In silico Tools for T Cell Epitope Prediction

As major histocompatibility complexes in humans, the human leukocyte antigens (HLAs) have important functions to present antigen peptides onto T-cell receptors for immunological recognition and responses.

Prediction the binding to HLA Class I (CD8+) or HLA Class II(CD4+) based on:

Structural model
Experimental data (machine learning algorithm)
Wide range of HLA alleles available

Reduce false-positive – Self-peptides:

Human antibody germline
Customized filter

Output

Epitope Identification and Ranking

Promiscuity (binding to multiple HLA allotypes)
Population frequency and importance of affected allotypes ( DR allotypes are the primary focus. DQ and DP have lower expression.)
Evaluation of the binding strength (strong/medium binders)
Impact of the TCR filters
Potential population impact (calculated by combining the population frequencies of the individual binding allotypes)
Overall RISK SCORE of the molecule ------ provides a ranking that can be compared to marketed antibodies

Available as

Detailed report
Server-based tool available for use in Customers' facilities
Heatmap for Protein Engineering and/or Deimmunisation

Cellular immune response is based on the specific recognition system mentioned above. The mechanism of peptide-MHC (pMHC) complexes recognized by T-cells is important for revealing cellular immunity process. T-cell-based immunotherapies against cancer can benefit from prediction of peptide-MHC binding affinity. CD ComputaBio provides in silico prediction services for researchers and pharmaceutical companies.

Services

Related Services:

TCR Binding Prediction