ROASMI: accelerating small molecule identification by repurposing retention data.

Area of Science:

• Analytical Chemistry
• Metabolomics
• Computational Chemistry

Background:

• Limited replicability of retention data in untargeted metabolomics hinders small molecule identification.
• Existing retention order models lack generalizability and predictive reliability.

Purpose of the Study:

• To present the ROASMI model for reliable prediction of retention order in reversed-phase liquid chromatography (RPLC).
• To improve small molecule identification in untargeted metabolomics by enhancing data reproducibility.

Main Methods:

• Coupling data-driven molecular representation with mechanistic insights to develop the ROASMI model.
• Validating ROASMI generalizability across 71 independent RPLC datasets.
• Applying ROASMI to real-world datasets for isomer differentiation and peak annotation.

Main Results:

• ROASMI demonstrates proven generalizability across diverse RPLC datasets.
• The model effectively distinguishes coexisting isomers with similar fragmentation patterns.
• ROASMI aids in annotating detection peaks lacking informative spectra.

Conclusions:

• ROASMI enables reliable retention order prediction, addressing key limitations in metabolomics.
• The model's flexibility allows retraining and compatibility with other MS/MS scorers for improved small molecule identification.