Deep learning model for predicting tumor-specific neoantigen mhc class i or class ii immunogenicity

A method for predicting tumor-specific neoantigen MHC class II immunogenicity, comprising:

• (a) obtaining a peptide sequence comprising a tumor-specific neoantigen and corresponding flanking regions;
• encoding the peptide sequence and the flanking regions into a numerical vector, wherein each numerical vector comprises amino acid residues encoding the peptide for the tumor-specific neoantigen and the amino acid residues of the flanking regions and the amino acid residue position;
• (b) obtaining a HLA allele pseudo-sequence and encoding the HLA allele pseudo-sequence into a numerical vector, wherein the HLA allele pseudo-sequence represents a HLA allele;
• (c) using a neural network model to jointly predict the tumor-specific neoantigen MHC class II binding affinity and a numerical probability for the tumor-specific neoantigen that a corresponding peptide will be presented by a MHC class II protein on a cell-surface;
• wherein the neural network model is trained by:
• (i) inputting training data set to optimize performance of the neural network model;
• wherein the training data set comprises peptide-MHC class II affinity measurement data sets and cell-surface peptide presentation data sets and an input layer comprising the numerical vector comprising the peptide sequence for the tumor-specific neoantigen and the flanking regions and the numerical vector comprising the HLA allele pseudo-sequence;
• (ii) encoding the numerical vector comprising the peptide sequence for the tumor-specific neoantigen and the flanking regions and the numerical vector comprising the HLA allele pseudo-sequence into an amino acid embedding layer;
• (iii) flattening the amino acid embedding layer to produce a numerical vector representation for each peptide sequence for the tumor-specific neoantigen and the peptide sequence flanking regions, and HLA allele pseudo-sequence;
• (iv) predicting the tumor-specific neoantigen MHC class II binding affinity by concatenating the tumor-specific neoantigen peptide sequence and the HLA allele pseudo-sequence, applying one or more layers and/or one or more activation functions, wherein the output is a numerical score representing the tumor-specific neoantigen MHC class II binding affinity;
• and
• (v) predicting a probability that the tumor-specific neoantigen will be presented by a MHC class II protein on a cell-surface by concatenating the peptide sequence of interest, the peptide sequence flanking regions, and the HLA allele pseudo-sequence into a single numerical vector, applying one or more layers and/or one or more activation functions, wherein an output is a numerical probability that a peptide will be presented by a MHC class II protein on a cell-surface;
• wherein the tumor-specific neoantigen MHC class II binding affinity and the numerical probability that the tumor-specific neoantigen will be presented by a MHC class II protein on a cell-surface is a proxy for tumor-specific neoantigen MHC class II immunogenicity.