TC-TopoRT: Tautomer-consistent and topology-aware retention time prediction for LC-MS metabolite annotation

Accurate retention time (RT) prediction provides chromatographic evidence complementary to precursor mass and MS/MS spectra in LC-MS metabolomics. Such evidence is particularly valuable for metabolite annotation, where multiple structurally plausible candidates may exhibit similar spectral signatures. Here, we present TC-TopoRT, a tautomer-consistent and topology-aware framework for RT prediction in LC-MS metabolite annotation. TC-TopoRT constructs paired molecular views from the dataset-provided SMILES and a conservative tautomer-standardized representation generated under identity-preserving checks. Each view is lifted into a ring-aware cell complex and encoded by a cellular Weisfeiler-Lehman network (CWN) to learn topology-aware molecular representations. The resulting paired-view predictions are then integrated through a leakage-free out-of-fold fusion protocol without using independent-test labels for model selection, calibration, or stacker fitting. On the retained-compound SMRT benchmark, TC-TopoRT achieved a mean absolute error of 25.06 s. A five-seed ensemble further reduced the error to 24.92 s. Across six PredRet chromatographic datasets, TC-TopoRT reported lower MAEs than the cited reference values on four systems and remained within 1.92 s on the other two. With RT thresholds calibrated on fixed development queries held out from calibration-model fitting and disjoint from final evaluation, TC-TopoRT reduced candidate spaces by 41.05% on MetaboBase and 30.35% on RIKEN-PlaSMA while retaining 42/45 (93.33%) and 84/85 (98.82%) true candidates, respectively. It also maintained or improved Top-k prioritization relative to the original MS-FINDER ranking. These results support the use of tautomer-consistent and topology-aware RT prediction as complementary chromatographic evidence for LC-MS metabolite annotation.