A Medicinal Chemistry-Centered Evaluation of AlphaFold 3 and Boltz-2 Across Diverse Binding Modalities
AlphaFold 3 (AF3) and Boltz-2 are state-of-the-art AI-based tools for biomolecular structure prediction, but whether their predictions provide useful guidance for lead optimization, SAR interpretation, and virtual screening remains insufficiently characterized. We benchmarked their performance using newly determined soluble epoxide hydrolase co-crystal structures and matched activity data…
AlphaFold 3 (AF3) and Boltz-2 are cutting-edge AI tools designed to predict the structures of biomolecules. However, it is unclear if the predictions from these tools can be relied upon for lead optimization, structure-activity relationship (SAR) interpretation, and virtual screening. To address this knowledge gap, researchers conducted a comprehensive benchmark study comparing the performance of AF3 and Boltz-2 against newly determined co-crystal structures of soluble epoxide hydrolase and associated activity data.
The study also incorporated a curated dataset containing information on kinases, allosteric modulators, covalent systems, PROTACs, molecular glues, fragments, membrane proteins, RNA binders, and activity-cliff pairs. Both AI models successfully captured canonical orthosteric enzyme and kinase complexes, including critical DFG/C conformational states.
Nevertheless, challenges arose when dealing with allosteric, membrane-protein, and induced-proximity complexes. The study found that pharmacophore Root Mean Square Deviation (RMSD) was frequently lower than the overall ligand RMSD, suggesting that key recognition features were preserved even when there were imperfect alignments of the entire ligand.
Notably, AF3's minimum Pose Accuracy Error (minPAE) showed a strong correlation with pose accuracy, while very low minPAE values were observed.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
