Identification of tumor-specific antigens (neoantigens) is needed for development of effective cancer immunotherapy and a good source for such antigens are the pools of HLA-bound peptides presented exclusively by the tumor cells. Mass spectrometry (MS) has evolved as the method of choice for the exploration of the human immunopeptidome (HLA class-I and class-II peptides). Workflows for immunopeptidomics are different from ones for more established shotgun proteomics, yet inherent differences between these two fields create significant drawbacks of current data analysis algorithms for the former. We provide AI-based solutions to address the barriers for data analysis, e.g., diverse C-termini of HLA-peptides, lack of sequence library for spliced peptides, no peptide de novo sequencing algorithms for data independent acquisition (DIA) method, etc. With deep learning technology, we proposed a new approach integrating motif-constrained database search and de novo sequencing for HLA-peptide identification that would increase peptide sequencing coverage, depth, and confidence, collectively enhancing the capabilities of the field of immunopeptidomics. Our approach was tested with several public data sets for the identification of HLA-peptides or neopeptides, including DDA (PXD007596, PXD006939, PXD004964, and PXD002431) and DIA (PXD001094). On average, 60% more HLA-peptides were identified by validation with the public immunopeptidome databases, IEDB and NeuroPep. Preliminary results showed that the AI-based data analysis would provide a novel solution for immunopeptidomics with high sensitivity and accuracy.
Reference