Optimizing Spectronaut’s AI-Based Peptide Property Predictors for Immunopeptidomic Applications

HLA-I and HLA-II molecules present peptides to T cells, playing a crucial role in immune surveillance, vaccine development, and immunotherapy, with significant implications for targeted cancer treatments. MS-based proteomics is essential for characterizing these peptides, enhancing our understanding of these processes.

AI-based peptide property prediction boosts the sensitivity of discovery proteomic workflows by providing reference coordinates for RT, CCS, and fragmentation. This is especially vital in large search spaces typical of immunopeptidomics.

Immunopeptides exhibit different physicochemical properties. These differences affect the distribution of peptide properties, necessitating adaptation of AI-based peptide property predictors for optimal performance in immunopeptidomic applications. Scaling laws in large language models (LLMs)2 suggest that increasing the amount of relevant data as much as possible helps AI models to generalize.