From Reverse Phase Chromatography to HILIC: Graph Transformers Power Method-Independent Machine Learning of Retention Times

Liquid chromatography (LC) is a cornerstone of analytical separations, but comparing the retention times (RTs) for different LC methods is difficult because of variations in experimental parameters such as column type and solvent gradient. Nevertheless, RTs are powerful metrics in tandem mass spectrometry (MS2) that can reduce false positive rates for metabolite annotation, differentiate isobaric species, and improve peptide identification. Here, we present Graphormer-RT, a novel graph transformer that performs the first “method-independent” prediction of RTs. We use the RepoRT dataset, containing 142,688 reverse phase (RP) RTs (191 methods) and 4,373 HILIC RTs (49 methods). Our best RP model achieved a test set mean average error (MAE) of 29.3±0.6 s, a significant improvement over the previous record (1 method). Our best performing HILIC model achieved a test MAE=42.4±2.9 s. Extending this proof-of-concept work could enable machine-optimization of automated LC workflows and in silico annotation of unknown analytes in LC-MS2 measurements.