Predicting Chemical Stressor Effects on Conjugative Plasmid Transfer Rates Using Machine Learning

The best-performing model combined concentration, Morgan fingerprints, and experimental conditions using extreme gradient boosting (XGB), achieving <i>Q</i>²ext = 0.373 on a held-out compound-aware test set and a median <i>Q</i>²ext of 0.319 within the defined applicability domain across 20 independent splits. SHAP analysis revealed that Morgan fingerprints account for 69% of total predictive weight, with quaternary ammonium compounds and long-chain alkyl structures associated with inhibition of conjugation and sulfonamide and amine-containing compounds associated with promotion.

These results demonstrate that chemical structural features provide critical information for predicting the chemical modulation of conjugative plasmid transfer. This work highlights that combining chemically informed descriptors with machine learning (ML) anticipates chemical contributions to antibiotic resistance dissemination, providing a scalable framework for the prioritization and risk assessment of chemicals.