Docking of millions: accelerating a million-scale virtual screening using deep learning.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Virtual screening is crucial for computer-aided drug discovery.
• Expanding chemical databases necessitate efficient screening methods.
• Accuracy and speed are key for ultra-large-scale virtual screening.

Purpose of the Study:

• To develop an ultra-large-scale virtual screening system integrating accurate binding affinity prediction and fast docking score estimation.
• To accelerate the drug discovery process by optimizing screening efficiency.

Main Methods:

• Integration of AK-Score2 for binding affinity prediction and V-Dock for docking scores.
• Iterative learning of V-Dock to approximate AK-Score2 affinities, enabling selective compound docking.
• Benchmarking on 5 million compounds against multiple targets (DDR1, c-kit, ASK1, NSD1, CREBBP, PDE5).

Main Results:

• Docking of Millions (DoM) reduced screening time to an average of 319 hours (12% of full screening).
• Achieved an 89% retrieval rate for top 100 compounds.
• Inhibition assays confirmed active molecules against ASK1 and DDR1, with IC50 values as low as 1.96 μM and 788 nM.

Conclusions:

• DoM provides a practical and efficient platform for ultra-large-scale virtual screening in drug discovery.
• The system significantly saves time and resources compared to traditional full library screening.
• Demonstrated success in identifying potent inhibitors for therapeutic targets.