Benchmarking Refined and Unrefined AlphaFold2 Structures for Hit Discovery.

Area of Science:

• Computational Biology
• Structural Biology
• Drug Discovery

Background:

• AlphaFold2 (AF2) accurately predicts protein 3D structures from amino acid sequences.
• The AlphaFold Protein Structure Database provides comprehensive human proteome coverage.
• Structure-based virtual screening is crucial for identifying drug candidates.

Purpose of the Study:

• To evaluate the virtual screening performance of AlphaFold2 (AF2) predicted protein structures.
• To assess the impact of structure refinement on AF2-based virtual screening.
• To determine the utility of AF2 structures for *in silico* hit identification.

Main Methods:

• Utilized the Glide molecular docking method for virtual screening.
• Compared AF2 structures with known *apo* and *holo* protein structures using the DUD-E dataset.
• Employed an induced-fit docking protocol (IFD-MD) for refining AF2 structures.

Main Results:

• AF2 structures showed comparable early enrichment to *apo* structures but lagged behind *holo* structures.
• Refinement of AF2 structures using IFD-MD improved virtual screening performance (avg. EF 1%: 18.9).
• Using Glide-generated poses as templates for IFD-MD also enhanced performance (avg. EF 1%: 18.0).

Conclusions:

• Refined AF2 structures demonstrate considerable promise for *in silico* hit identification in drug discovery.
• Proper preparation and refinement are key to maximizing the utility of AF2 structures for virtual screening.
• AF2-derived structures can be valuable assets in early-stage drug discovery pipelines.