Suitability of AF2-derived decoy structures for docking and virtual screening

Determine the suitability of decoy protein structures generated by AlphaFold2-based ensemble diversification methods—specifically reduced multiple sequence alignment AlphaFold2 (rMSA AF2), AF2-cluster, and AlphaFlow—for use as receptor models in downstream docking and virtual screening workflows.

Background

The paper surveys several AlphaFold2-based approaches that increase structural diversity by generating decoy conformations (rMSA AF2, AF2-cluster, AlphaFlow). Despite producing diverse structures away from the native state, the authors emphasize that it is not yet established whether such decoys are appropriate receptor models for docking or virtual screening, a central step in computer-aided drug design.

This uncertainty directly impacts practical deployment of AF2-generated ensembles for ligand discovery, motivating methods like AF2RAVE to physically rank and select decoys with Boltzmann weights before docking.

References

Several AF2-based techniques, including reduced multiple sequence alignment (rMSA) AF2 (or MSA subsampling AF2), AF2-cluster and AlphaFlow, have been devised to generate distinct decoy structures from native states. However, the suitability of these decoys for subsequent docking and virtual screening remains uncertain.

— Empowering AlphaFold2 for protein conformation selective drug discovery with AlphaFold2-RAVE  (2404.07102 - Gu et al., 2024) in Introduction (Section 1)