Benchmarking Spectral Library Prediction Platforms for Neuropeptidomics Applications

Data-independent acquisition (DIA) mass spectrometry has emerged as a powerful tool for neuropeptidomics, but its success relies heavily on the quality of spectral libraries used for peptide identification. There are inherent challenges to mass spectrometry analysis of crustacean neuropeptides, including the endogenous nature in which they are analyzed, extensive post-translational modification (PTM), and atypical fragmentation patterns. Thus, general-purpose proteomic spectral prediction tools may not perform optimally in the endogenous peptide domain. In this study, we benchmark four widely used spectral prediction platforms, Prosit, MS2PIP, AlphaPeptDeep, and UniSpec, to evaluate their performance in predicting the fragmentation of neuropeptides. Using an empirically derived spectral library from crustacean tissues as reference, we assess model compatibility, dot-product similarity, Pearson correlation, and DIA-based identifications across brain, sinus gland, and pericardial organ samples. Our results reveal that no single model comprehensively captures neuropeptide fragmentation characteristics. While UniSpec showed unexpected strengths due to its inclusion of neutral loss ions, AlphaPeptDeep demonstrated the highest spectral similarity, and MS2PIP and Prosit outperformed in DIA-NN identifications. We further highlight the critical impact of neutral loss fragments, present in over 50% of empirical spectra, and emphasize the need for hybrid spectral libraries that integrate complementary strengths across models. This work provides a foundational framework for optimizing spectral library selection in neuropeptidomics and underscores the importance of model-specific biases when analyzing structurally diverse endogenous peptides.