DGMM: A Deep Learning-Genetic Algorithm Framework for Efficient Lead Optimization in Drug Discovery.

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Lead optimization in drug discovery requires balancing structural diversity with core molecular features and optimizing biological activity against drug-like properties.
• Existing methods face challenges in efficiently exploring chemical space while maintaining scaffold integrity and desired pharmacological profiles.

Purpose of the Study:

• To introduce and validate the Deep Genetic Molecule Modification (DGMM) algorithm, a novel computational framework for efficient molecular optimization in drug discovery.
• To demonstrate DGMM's capability in balancing structural variation with scaffold retention and optimizing drug likeness against target activity.

Main Methods:

• Developed the Deep Genetic Molecule Modification (DGMM) algorithm, integrating deep learning (variational autoencoder with scaffold constraints) and genetic algorithms.
• Employed a multiobjective optimization strategy using Monte Carlo search and Markov processes for trade-off exploration.
• Validated DGMM retrospectively on CHK1, CDK2, and HDAC8 targets, followed by a prospective campaign for ROCK2 inhibitor discovery.

Main Results:

• DGMM demonstrated state-of-the-art performance in activity optimization, generating structurally diverse yet pharmacologically relevant compounds.
• Retrospective validation successfully reproduced known optimization pathways for selected targets.
• Prospective campaign led to the discovery of novel ROCK2 inhibitors with a 100-fold increase in biological activity.

Conclusions:

• The DGMM algorithm is an effective computational tool for structural optimization of drug molecules.
• DGMM successfully balances structural diversity, scaffold retention, and optimization of biological activity and drug-like properties.
• The prospective discovery of potent ROCK2 inhibitors validates DGMM's utility in real-world drug discovery campaigns.