An artificial intelligence accelerated virtual screening platform for drug discovery.

Area of Science:

• Computational chemistry and drug discovery
• Structural biology
• Artificial intelligence in pharmacology

Background:

• Structure-based virtual screening (SBVS) is crucial for identifying drug candidates.
• Accurate prediction of binding pose and affinity is essential for SBVS success.
• Current methods face challenges with large chemical libraries and receptor flexibility.

Purpose of the Study:

• To develop a highly accurate SBVS method, RosettaVS, for predicting docking poses and binding affinities.
• To create an open-source, AI-accelerated platform for large-scale virtual screening.
• To identify novel hit compounds against therapeutic targets KLHDC2 and Na_V1.7.

Main Methods:

• Developed RosettaVS, an advanced structure-based virtual screening method.
• Incorporated receptor flexibility modeling into the screening approach.
• Utilized an AI-accelerated platform to screen multi-billion compound libraries against KLHDC2 and Na_V1.7.

Main Results:

• RosettaVS demonstrated superior performance over state-of-the-art methods on benchmarks.
• Identified seven hit compounds (14% hit rate) for KLHDC2 and four hits (44% hit rate) for Na_V1.7.
• All identified hits exhibited single-digit micromolar binding affinities and were discovered in under seven days.

Conclusions:

• The AI-accelerated RosettaVS platform enables rapid and accurate virtual screening of massive compound libraries.
• The method successfully identified novel lead compounds for challenging targets.
• X-ray crystallography validated the predicted binding pose of a KLHDC2 ligand, confirming the method's efficacy in lead discovery.