Can We Reliably Identify Drug Targets at the Proteome Scale Using AlphaFold based Docking or Generative AI?

Drug target identification is important for drug repurposing, follow-up studies in drug development, and understanding drug side effects. Essentially, the pharmacological effect of drug is determined by the interaction between the drug and its target, and the interaction is in turn governed by the three-dimensional structures. With the advent of AlphaFold, we now have access to nearly complete proteome-scale structural data for any species. This opens the possibility of systematically identifying drug targets in silico using current docking and modeling tools. In this study, we benchmarked the performance of several docking tools in proteome-wide blind reverse docking and also evaluated the utility of generative AI, including AlphaFold3, in drug target prediction. Our results show that, with appropriate tool selection and preparation of the structural dataset, drug targets can be effectively enriched. Notably, AlphaFold3 alone enables reasonably accurate identification of potential targets. As a proof-of-concept, we identified a previously unreported binding target for the clinical drug palbociclib. This work provides practical insight into how we can utilize structural data for drug target discovery in the post-AlphaFold era.