RAPID-Net: Accurate Pocket Identification for Binding-Site-Agnostic Docking.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Accurate identification of druggable pockets is crucial for structure-based drug design and docking.
• Existing methods may not fully capture pocket features or integrate seamlessly with docking pipelines.

Purpose of the Study:

• To present RAPID-Net, a deep learning algorithm for precise prediction of binding pockets.
• To evaluate RAPID-Net's performance in structure-based drug design and virtual screening.
• To demonstrate RAPID-Net's capability in identifying novel binding sites for therapeutic development.

Main Methods:

• Development of RAPID-Net, a deep learning-based algorithm for pocket prediction.
• Integration of RAPID-Net with AutoDock Vina for docking simulations.
• Performance evaluation on the PoseBusters benchmark and diverse datasets.
• Comparison with existing pocket prediction tools (DiffBindFR, PUResNet, Kalasanty) and AlphaFold 3.

Main Results:

• RAPID-Net-guided AutoDock Vina achieved 54.9% Top-1 poses with RMSD < 2 Å on PoseBusters.
• On challenging data, RAPID-Net-guided Vina achieved 53.1% Top-1 poses (vs. 59.5% for AlphaFold 3).
• RAPID-Net identified at least one pose with RMSD < 2 Å in 92.2% of cases, highlighting pose ranking as a bottleneck.
• RAPID-Net outperformed other tools in docking accuracy and pocket-ligand intersection rates.
• Accurate identification of distal functional sites and broader pocket identification for SARS-CoV-2 RdRp.

Conclusions:

• RAPID-Net offers accurate binding pocket prediction, enhancing docking pipeline integration.
• Its lightweight inference, scalability, and competitive accuracy make it suitable for large-scale virtual screening.
• RAPID-Net facilitates the discovery of novel therapeutic targets, including allosteric sites.
• The algorithm shows potential for accelerating novel therapeutic development by uncovering diverse binding pockets.