Abstract 2340: STRUMP-I: Structure-based binding prediction of MHC class I

Abstract The presentation of short peptides derived from endogenously expressed proteins within the Major Histocompatability Complex-I (MHC-I), together known as the pMHC, allows the adaptive immune system to monitor the internal functioning of a cell and surveil for foreign or mutated self-proteins. With the rising importance of immunotherapies targeting neoantigens in cancers, the ability to accurately predict which peptides will bind to the diverse population of MHC alleles is critically important. Computational methods for predicting MHC/peptide interactions fall into two broad categories: sequence-based methods that utilize machine learning or deep learning to predict binding from information about the MHC and peptide sequence alone, and structure-based methods that leverage the computed structure and energetics of the pMHC to predict binding. While sequence-based methods have outperformed structure-based methods, their performance varies with the quantity and quality of training data for each MHC allele. The pMHC structure also provides additional insights into downstream analyses, such as predicting the immunogenicity of cancer neoantigens, that cannot be determined from the sequence alone. We developed a STRUMP-I (STRUcture-based pMHC Prediction (for class I)), a novel pMHC binding prediction tool that combines structural modelling with a machine learning-based classifier to discriminate binders from non-binders. STRUMP-I has comparable performance to the state-of-the-art sequence-based methods on data taken from IEDB, the HLAthena training set, TESLA, PRIME, and an in-house cancer neoantigen dataset. However, it generalizes better to alleles with low representation or a lower proportion of positive examples in the training data than the sequence-based methods. Additionally, we demonstrate how STRUMP-I can be used in conjunction with PRIME, a sequence-based immunogenicity prediction method, to use structural information about the peptide to filter out false positive predictions and increase the precision of immunogenicity predictions, increasing the precision by up to 2-fold. Further, on a set of 10 experimentally validated peptides from the in-house cancer neoantigen dataset containing only a single immunogenic peptide, STRUMP-I filtered all false positive predictions and retained only the true positive prediction. By leveraging structural information about the pMHC, STRUMP-I provides robust binding predicting performance across MHC alleles and increases the precision of immunogenicity prediction for neoantigen screening. Citation Format: Adam Voshall, Woongyang Park, Eunjung Alice Lee, Yoonjoo Choi. STRUMP-I: Structure-based binding prediction of MHC class I [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 2340.