Enabling target-aware molecule generation to follow multi objectives with Pareto MCTS

Area of Science:

• Computational chemistry and cheminformatics
• Artificial intelligence in drug discovery
• Molecular modeling and simulation

Background:

• Target-aware drug discovery accelerates the design of small-molecule ligands with high binding affinity.
• Current deep generative models focus on protein-ligand binding but neglect other crucial drug properties like drug-likeness.
• A gap exists in multi-objective optimization for target-aware molecule generation using deep learning.

Purpose of the Study:

• To introduce ParetoDrug, a Pareto Monte Carlo Tree Search (MCTS) algorithm for multi-objective target-aware molecule generation.
• To enable synchronous optimization of multiple molecular properties, including binding affinity and drug-likeness.
• To address limitations of existing methods by incorporating drug-likeness alongside binding affinity.

Main Methods:

• Developed ParetoDrug, a Pareto MCTS generation algorithm for chemical space exploration.
• Utilized pretrained atom-by-atom autoregressive generative models for guided molecule generation within MCTS.
• Introduced the ParetoPUCT scheme to balance exploration and exploitation during atom selection.

Main Results:

• ParetoDrug effectively navigates complex chemical spaces to discover novel compounds.
• The algorithm successfully identifies molecules with satisfactory binding affinities and drug-like properties.
• Demonstrated high effectiveness across various multi-objective target-aware drug discovery tasks through benchmark experiments and case studies.

Conclusions:

• ParetoDrug bridges the gap in multi-objective target-aware molecule generation.
• The proposed method enables simultaneous optimization of binding affinity and drug-likeness.
• ParetoDrug represents a significant advancement in deep learning-based drug discovery for generating effective drug candidates.