Deep Learning Methods For Prediction of HLA-Peptide Interactions in IDB

Inflammatory bowel disease (IBD) is a complicated gastrointestinal disease linked to so many factors. Studying the interactions and functions of proteins in diseases is a challenging task. Here, we tried to understand the relationship between Human Leukocyte Antigen (HLA) genes and peptides in the development of IBD. Two types of HLA genes, HLA DRB1* 13:01 and HLA DRB1*15:01 were considered which are highly important for presenting peptides to T-cells as one of their immunological activities, therefore, it is necessary to study their interaction with peptides. An improved version of deep learning models based on convolutional, recurrent neural networks and a hybrid of both models were proposed in this study by utilizing an embedding matrix as part of the model architecture for better encoding of amino acid features, this technique has proven to be highly effective when compared to the state of the art models. The different deep model architectures were trained and tested to predict HLA DRB1*13:01 and HLA DRB1*15:01 peptide interactions. The hybrid method gave the best performance with 0.987 as the Spearman correlation. The models proposed in this work will give better insight into the interaction between peptides and HLA genes in the development of IBD.