Combining Sequence Similarity with Physicochemical Properties to Predict Binders for MHC-II Molecules

Abstract

Prediction of binders of the major histocompatibility complex class II (MHC-II) molecules is critical for T cell immunogenicity. As protein-protein interaction also relies on physicochemical properties, we aim to build a novel model combining sequence information and the physicochemical properties of proteins. Our research used data from the NetMHCIIpan 3.2 study. Features include BLOSUM50 and the physicochemical properties from iFeature Python package. We created a hybrid model of recurrent neural layers and feedforward layers. The final Area Under the Receiver Operating Characteristics (AUROC) on the test data was 0.755.