A review in Eur J Med Chem (17 October 2026) takes stock of AlphaFold structures in drug discovery. It covers target identification, binding-site prediction, docking, virtual screening, lead optimization and SAR interpretation, with examples in oncology, antimicrobial resistance, neurodegeneration and targeted protein degradation.
Two numbers stand out. For the GPCR TAAR1, docking against AlphaFold2 models gave a 60% hit rate versus 22% for a homology model. For the kinase CDK20, a workflow combining an AlphaFold model with generative chemistry produced a compound with an IC50 of 33.4 nM and roughly eightfold antiproliferative selectivity between cell lines.
The most useful part for a medicinal chemist is the evidence-graded decision framework: it states when a predicted structure is likely sufficient for hit finding, SAR interpretation, selectivity optimization, fragment-based design and degrader design, and where quantitative evidence is still missing.
The limits remain clear: protein flexibility, ligand-induced conformational change, binding-affinity prediction and biological complexity. The authors stress that predicted structures still need experimental validation before they guide synthesis.
Takeaway: an AlphaFold model is a good starting point for hit finding, not a replacement for crystallography when you are optimizing selectivity or affinity.
Reference: 10.1016/j.ejmech.2026.119442, Eur J Med Chem, 17 Oct 2026 (PubMed).