FragOPT: An ML-Driven Computational Workflow for Rational Fragments Optimization Toward Lead Compounds.

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Machine learning (ML) accelerates drug discovery but faces challenges due to chemical space complexity and "black box" models.
• Predicting drug-target interactions and properties is crucial but often limited by current computational methods.

Purpose of the Study:

• To introduce FragOPT, a comprehensive workflow for optimizing molecules against specific protein targets.
• To enhance drug discovery efficiency and precision by integrating ML interpretability into molecular design.

Main Methods:

• FragOPT utilizes a classification model and SHAP (SHapley Additive exPlanations) for identifying bioactive molecular fragments.
• Disadvantageous fragments are redesigned using deep learning, then recombined with advantageous fragments for enhanced binding affinity.
• The workflow was validated on targets relevant to solid tumors and SARS-CoV-2.

Main Results:

• FragOPT-generated molecules showed superior synthesizability and binding affinity compared to other fragment-based methods.
• Molecular mechanics with generalized Born and surface area (MM/GBSA) and free energy perturbation (FEP) calculations confirmed lower binding free energies for novel molecules.
• The method demonstrated significant optimization of the initial drug discovery process.

Conclusions:

• Integrating Quantitative Structure-Activity Relationship (QSAR) models and interpretability methods optimizes molecular generation.
• FragOPT offers a precise and efficient pathway for developing novel therapeutics.
• The FragOPT workflow is publicly available for use in drug discovery research.