Predicting HLA ligands through a ubiquitination-aware random forest

Abstract

In the realm of immunotherapy, identification of human leukocyte antigen (HLA) ligands has an essential role. In the research described in this paper, we present a novel approach to predict HLA ligands by integrating the post-translational modification ubiquitination. The methodology involves the development of a random forest algorithm leveraging a combined input of ubiquitination data and results from existing prediction method netMHCpan. Ubiquitination-aware random forest performance was evaluated by benchmarking against existing prediction methods. Marginal performance enhancement of the method is observed in two distinct datasets; illustrating how ubiquitination might play a small role in modulating HLA ligand presentation, thereby influencing immune recognition. Therefore, the results hint that including ubiquitination in predictive methods may slightly improve identification of HLA ligands, but further validation is required. More surprisingly, our method performed well at HLA ligand prediction on the validation cell line, despite differing HLA subtype expression, indicating an unexpected wide applicability.