ReBADD-SE: Multi-objective molecular optimisation using SELFIES fragment and off-policy self-critical sequence...

Area of Science:

• Medicinal Chemistry
• Computational Drug Design
• Artificial Intelligence in Drug Discovery

Background:

• Developing drugs to selectively eliminate diseased cells is a significant challenge in computer-aided drug design.
• Existing multi-objective molecular generation methods excel at kinase inhibitor tasks but may not effectively generate molecules violating Lipinski's rule of five, like navitoclax.
• The limitations of current models in generating rule-violating molecules necessitate novel approaches.

Purpose of the Study:

• To address the limitations of existing methods in generating molecules that violate Lipinski's rule of five.
• To propose an advanced multi-objective molecular generation method capable of designing complex drug candidates.
• To enhance the efficiency and effectiveness of multi-objective molecular optimization in drug discovery.

Main Methods:

• A novel parsing algorithm for molecular string representation was developed.
• A modified reinforcement learning approach was implemented for efficient multi-objective molecular optimization training.
• The proposed method was evaluated on inhibitor generation tasks, including GSK3b+JNK3 and Bcl-2 family targets.

Main Results:

• The developed model achieved an 84% success rate in generating GSK3b+JNK3 inhibitors.
• The model demonstrated a 99% success rate in generating Bcl-2 family inhibitors.
• The method shows promise in generating molecules with specific properties, including those violating Lipinski's rule.

Conclusions:

• The proposed multi-objective molecular generation method effectively addresses limitations of existing approaches.
• The novel parsing and reinforcement learning techniques enable efficient optimization for complex molecular generation.
• This advancement holds significant potential for computer-aided drug design, particularly for generating disease-targeting therapeutics.