This approach uses AlphaFold2 (an AI that predicts 3D protein shapes) not just to get one structure per protein, but to explore many plausible shapes of a drug target. These AI‑generated shapes are then used as ‘locks’ in large-scale virtual screening to find small‑molecule ‘keys’ (drug candidates) that fit, even when proteins flex or change shape.
Traditional virtual screening often fails when only a single static protein structure is available, especially if the biologically relevant binding conformations are unknown or underrepresented. This method systematically explores the structural space predicted by AlphaFold2 to generate multiple realistic target conformations, improving hit rates and relevance in structure-based virtual screening and de‑risking early-stage drug discovery when experimental structures are missing or incomplete.
Combination of AlphaFold2-based conformational sampling workflows with in‑house structural biology knowledge, proprietary target selection, and historical virtual screening and assay data can form a defensible loop that continuously improves screening performance and is hard for competitors to replicate quickly.
Hybrid
Unknown
High (Custom Models/Infra)
Compute cost and throughput for generating and managing large ensembles of AlphaFold2-derived protein conformations, and then docking or virtually screening large libraries against each conformation (GPU/CPU time, storage, and I/O).
Early Adopters
Unlike simple use of a single AlphaFold2 prediction per target, this work explicitly explores the structural space around AlphaFold2 outputs to produce multiple conformers tailored for structure-based virtual screening, effectively turning AlphaFold2 into a generator of target ensembles rather than a one-shot structure predictor.
5 use cases in this application