DeepHLApan: A Deep Learning Approach for High-Confidence Neoantigen Prediction

Background:  Neoantigens are the most widely recognized elements to distinguish cancer and normal cells and consequently play important roles in cancer immunotherapy. Current methods used for neoantigen prediction focus on the binding between human leukocyte antigens (HLAs) and peptides, which is insufficient for high-confidence neoantigen prediction.   Methods:  We apply deep learning techniques to predict neoantigens considering both the possibility of mutant peptide presentation (binding model) and the potential immunogenicity (immunogenicity model) of the peptide-HLA complex (pHLA) present on the cell surface.   Findings:  The binding model achieves performance comparable to or even better than that of other well-acknowledged tools with the latest Immune Epitope Database (IEDB) benchmark datasets. Using the immunogenicity model, we demonstrate that limited immunogenicity data could significantly improve the identification of high-confidence neoantigens. We further apply our method to mutations with pre-existing T-cell responses and ranked most of them (69%) in the top 20 under an expression threshold of transcripts per million (TPM)>2.   Interpretation:  The process of neoantigens inducing T cell response is complex and the immunogenicity of pHLA should be considered for high-confidence neoantigen prediction. Funding Statement: This work has been supported by the National Key R&D Program of China (Grant No. 437 2017YFC0908600), the Zhejiang Provincial Natural Science Foundation of China 438 (Grant No. LY19H300003), and the Fundamental Research Funds for the Central 439 Universities of China. Declaration of Interests: The authors declare that they have no competing interests. Ethics Approval Statement: Not needed.