Mining for Potent Inhibitors through Artificial Intelligence and Physics: A Unified Methodology for Ligand Based and...

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Drug discovery is a lengthy and costly process.
• Computer-aided methods are crucial for accelerating drug discovery.
• Developing novel molecules with desired properties is challenging.

Purpose of the Study:

• To develop a machine learning algorithm for generating novel drug molecules.
• To combine deep reinforcement learning with molecular docking for efficient drug design.
• To create a versatile platform adaptable to various target proteins.

Main Methods:

• Developed the iMiner algorithm integrating deep reinforcement learning and AutoDock Vina.
• Employed diverse reward functions for chemical novelty and target interaction.
• Integrated a workflow for filtering compounds (PAINS, Lipinski violations, etc.).
• Included options for cross-validation and molecular dynamics simulations.

Main Results:

• iMiner generates novel inhibitor molecules with shape and compatibility constraints.
• The algorithm successfully incorporates chemical similarity, fragment-based growth, and residue interactions.
• The workflow effectively filters undesirable compounds and assesses pose stability.
• The approach is protein-structure-dependent, allowing broad applicability.

Conclusions:

• iMiner offers a powerful, adaptable approach for rapid drug discovery.
• The algorithm accelerates the identification of novel small molecule therapeutics.
• This method can be applied to any target protein for inhibitor development.