Structure-based retention time prediction for fentanyl analogs under cyano and phenyl-hexyl SCC-LC–MS/MS conditions: supporting discrimination among MS-ambiguous candidates

Isomeric and isobaric fentanyl analogs can share precursor- and product-ion signals, limiting candidate discrimination by mass spectrometry (MS) in non-targeted screening. Retention time (RT) provides orthogonal chromatographic information, but measured reference RTs are unavailable for many candidate structures. We therefore developed a structure-based RT predictor for fentanyl analogs analyzed by liquid chromatography–tandem mass spectrometry using serially coupled cyano and phenyl-hexyl columns (SCC-LC-MS/MS). Ninety-four curated structure–RT records were represented by 75 selected Mordred descriptors and paired with five regression algorithms. The resulting pipelines were evaluated using repeated nested cross-validation (CV), leave-one-out cross-validation (LOOCV), y-scrambling, and k-nearest-neighbor applicability domain (AD) analysis. Mean CV R 2 values were closely clustered (0.887–0.906), with the Mordred × Gaussian process regression pipeline (radial basis function kernel) ranking highest. In LOOCV, the selected pipeline achieved a mean absolute error (MAE) of 1.15 min, and 85.1% of records were predicted within ± 2.0 min of measured RT; the AD analysis classified 94.7% of records as in-domain. Within 23 MS-ambiguous candidate sets, predicted RTs correctly ordered 86% of pairs separated by more than 1.08 min (the subset MAE), compared with 68% of more closely spaced pairs. The largest prediction errors were systematic, concentrated on analogs whose distinctive structural modification was sparsely represented in the training data, indicating which additional reference standards would most extend reliable coverage. These internal validation results support the use of predicted RT as auxiliary chromatographic evidence for prioritizing MS-ambiguous fentanyl-analog candidates under the represented SCC-LC-MS/MS conditions.