Saturday 1 August 2020

(internal) neoIM algorithm *ongoing*

neoIM - peptide immunogenicity prediction

Only a very small fraction of putative neoantigens arising from a patient’s mutanome are actually able to evoke an immune response. This is due to the many processing steps necessary for a neoantigen to achieve T cell recognition (Figure 1). Hence, the mixed results in clinical attempts at utilising neoantigens for immunotherapy can mostly be attributed to the difficulty of finding truly immunogenic peptides from the set of novel peptides generated by a patient’s mutations.

Figure 1. An overview of the different neoantigen processing steps necessary for T cell recognition.

This low rate of actionable neoantigens causes high time and resource costs for the development of patient-tailored immunotherapies, as a large amount of candidate neoantigens need to be screened in order to obtain a small set of biologically relevant ones. In silico approaches can help alleviate this heavy cost by reducing the neoantigen search space,

The neoIM algorithm was developed as a tool to assess whether a neoantigen that is predicted to be presented on the cell surface is actually likely to be recognised by T cells. The algorithm was trained using a large dataset of experimentally validated immunogenic peptide sequences as well as a negative dataset consisting of MHC-presented non-immunogenic peptides. The model was then trained to distinguish between immunogenic and non-immunogenic peptides based solely on the peptide’s physicochemical properties (Figure 2).

Figure 2. The neoIM peptide immunogenicity prediction algorithm is trained on features extracted from a large dataset of immunogenic and non-immunogenic peptides.

The neoIM algorithm has shown its ability to identify immunogenic peptides on multiple public validation datasets. Additionally, immunogenicity assays were performed to validate neoIM predictions for peptides originating from various mutation types (SNV, frameshift Indel, Gene Fusion).

The application of neoIM can lower the false positive rate of peptide immunogenicity predictions significantly and allow for an optimal selection of neoantigens before clinical validation in any therapy context. As such, it represents a cost-efficient preliminary step in the search for actionable, immunogenic neoantigens and can deliver valuable insights on the properties of immunogenic peptides

Pfitzer et al., “neoIM: a machine learning model for T cell mediated immunogenicity prediction of MHC class I restricted epitopes”, in preparation.