Beyond rigid docking: deep learning approaches for fully flexible protein-ligand interactions.

Area of Science:

• Computational chemistry
• Biophysics
• Drug discovery

Background:

• Molecular docking predicts protein-ligand interactions, crucial for drug discovery.
• Traditional methods are computationally intensive, often sacrificing accuracy for speed in virtual screening.
• Deep learning (DL) models show promise in enhancing molecular docking accuracy and efficiency.

Purpose of the Study:

• To review the impact of DL on molecular docking.
• To examine current challenges and emerging solutions in DL-based docking.
• To explore future directions for improving computational predictions of biomolecular interactions.

Main Methods:

• Review of recent advancements in DL for molecular docking.
• Analysis of DL model performance compared to traditional docking algorithms.
• Exploration of techniques incorporating protein flexibility into DL docking models.

Main Results:

• DL significantly reduces computational costs while maintaining or improving docking accuracy.
• DL models face challenges in generalization and predicting accurate molecular properties (stereochemistry, bond lengths).
• Incorporating protein flexibility into DL models shows potential for more realistic biomolecular interaction predictions.

Conclusions:

• DL has transformed molecular docking, offering a powerful alternative to traditional methods.
• Addressing DL model limitations, such as generalization and physical realism, is crucial.
• Future research focusing on protein flexibility and advanced DL architectures will further enhance drug discovery pipelines.