LipiDetective: a deep learning model for the identification of molecular lipid species in tandem mass spectra.

Abstract

Confidently identifying lipids in samples is a prerequisite for understanding their many functions in health and disease. However, accurate molecular lipid species identification via tandem mass spectrometry remains challenging. Most current approaches match measured spectra against an in-house reference library, which hinders the comparability of results. To address this challenge, the transformer model LipiDetective was developed and trained on a dataset of spectra from lipid standards, databases, and publications. Learning the characteristic lipid fragmentation patterns, LipiDetective can accurately annotate molecular lipid species in tandem mass spectra independently of the experimental setup. Integrated gradients reveals that LipiDetective focuses on peaks matching known fragments, making its predictions humanly interpretable. Therefore, LipiDetective offers a data-driven approach for molecular lipid species identification that may improve the comparability of annotations across different laboratories and analysis workflows.