Publications

Identifying the molecular targets of toxic compounds remains a major challenge in toxicology, particularly when adverse effects occur in off-target organs and the mechanism of action is unknown. To address this issue, a comprehensive computational pipeline was developed to perform high-throughput molecular docking across the entire AlphaFold2-predicted structural proteome of representative organisms such as human and mouse, followed by enrichment analysis to estimate biological processes potentially affected by ligand binding. The pipeline was first evaluated using six known drug–target pairs. In several cases, the known targets were ranked between the top 2 and 250 proteins (top 0.009–1.15%) among more than 21,000 proteins, and displayed docking poses consistent with experimentally observed binding conformations. However, performance was limited for certain targets, such as carbonic anhydrase II with acetazolamide, where the binding pocket was broad, leading to inaccurate docking results. The pipeline was subsequently applied to puberulic acid, a compound suspected of causing severe nephrotoxicity. Screening identified sodium/myo-inositol cotransporter 2 (SLC5A11) as a high-affinity target in both human and mouse, suggesting a mechanism involving disruption of renal osmoregulation. Although docking scores represent only theoretical binding estimates and do not directly imply physiological effects, their distribution was independent of protein length and AlphaFold2 confidence scores (pLDDT), supporting the methodological robustness. This in silico framework enables hypothesis-driven identification of potential target proteins for toxicants or therapeutics and offers a useful tool for predictive toxicology, particularly when experimental data are limited. The pipeline is available at: https://github.com/toxtoxcat/reAlldock.