Leveraging Consensus Docking Approaches for Human Mitochondrial Complexes I and III

Although recent progress has been made, structure-based methods such as molecular docking are still underexplored in the context of toxicity prediction. These approaches offer added value, particularly in addressing challenges such as activity cliffs─i.e., caused by stereoisomerism─that are difficult to capture by conventional Quantitative Structure-Activity Relationship (QSAR) methods. In this study, we investigated the ability of docking scoring functions and protein-ligand interaction fingerprints to rank the potential hazard of compounds targeting the human mitochondrial complexes I and III (CI, NADH:ubiquinone oxidoreductase and CIII, cytochrome bc₁ complex). We applied an induced fit docking protocol to account for binding site flexibility and performed a set of binding energy minimizations for rescoring of representative binding modes. Both individual scoring functions and consensus scoring approaches achieved acceptable rank correlation to experimentally derived data from CIII (Spearman r: 0.89 and 0.86). Moreover, consensus interaction fingerprints that combine molecular interactions from both docking outputs captured differences of inhibitor subtypes at CIII. Follow-up in vitro testing confirmed an isomerism-dependent activity cliff of E-/Z-Fenpyroximate at CI. These findings support the utility of using consensus docking and scoring as a screening-level tool for prioritizing compounds based on interpretable predicted relative binding affinities at CI and CIII.