Mothra: Multiobjective de novo Molecular Generation Using Monte Carlo Tree Search.

Area of Science:

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

Background:

• Drug discovery requires identifying compounds with multiple optimal properties, a complex challenge due to vast chemical space.
• Existing multiobjective optimization methods often use linear combinations, which can oversimplify complex relationships and introduce weighting issues.
• Generative models for drug discovery need scalable solutions that handle multiple objectives effectively.

Purpose of the Study:

• To develop a scalable multiobjective molecular generative model using deep learning for drug discovery.
• To overcome the limitations of linear combination approaches in multiobjective optimization for molecular generation.
• To create a framework that integrates target affinity, drug similarity, and toxicity for enhanced compound generation.

Main Methods:

• Integration of recurrent neural networks (RNNs) for molecular generation.
• Application of Pareto multiobjective Monte Carlo tree search (MCTS) for determining optimal search directions.
• Development of enhanced evaluation functions incorporating target protein affinity, drug similarity, and toxicity.

Main Results:

• The proposed model demonstrates effectiveness in generating compounds that satisfy multiple criteria.
• Experimental results show significant improvements in key evaluation metrics compared to existing methods.
• The model successfully addresses the limitations associated with linear combination strategies.

Conclusions:

• The developed deep learning model offers a powerful and scalable approach for multiobjective optimization in drug discovery.
• This method enhances the efficiency and effectiveness of identifying promising drug candidates.
• The open-source release of the Mothra model and associated tools facilitates broader research and application in the field.