Dataset-specific retention time adjustment and multi-evidence lipid annotation: a reproducible R workflow for untargeted LC–MS lipidomics

Background and Objective: Untargeted lipid annotation by liquid chromatography-mass spectrometry is challenged by structural isomers, multiple ion forms, and retention time differences between analytical datasets. We developed an open-source computational workflow combining dataset-specific retention time adjustment with experimentally characterized lipid information to restrict candidate assignments while preserving unresolved identities. Methods: The R workflow uses a curated database of 592 lipid species derived from a human plasma reference material. It combines monotonic retention time adjustment, accurate-mass matching, isotope-pattern assessment, empirical adduct profile, and ambiguity resolution. Developed using a 40-sample MICROSAMPLING study, it was applied to an independent 189-sample METFORMIN-HIIE study acquired under compatible chromatographic conditions. Annotation stages were evaluated against fixed workflow-derived references, including a comparison with unadjusted database retention times. Results: Unadjusted retention time matching recovered 55.4-77.2% of reference feature-lipid pairs, whereas adjusted matching recovered all reference pairs under identical tolerances. RT adjustment reduced candidates on non-reference features by 69.5-83.2% relative to mass-only matching and increased within-reference candidate discrimination from 69.1-70.9% to 88.5-92.6%. Isotope filtering and ambiguity resolution further restricted assignments. Integration across ionization modes yielded 433 and 379 endogenous feature records, respectively; 93 and 100 retained multiple compatible lipid identities rather than forcing unsupported unique assignments. Conclusions: Dataset-specific adjustment improves agreement with workflow-curated assignments while restricting candidate space. The approach supports reproducible lipid annotation and transparent uncertainty handling. Because the reference assignments originate from the workflow itself, complete retention does not establish independent annotation sensitivity or accuracy.