Selective Inhibitor Design for Kinase Homologs Using Multiobjective Monte Carlo Tree Search

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery

Background:

• Designing selective molecules for drug targets is a complex, multiobjective challenge.
• Artificial intelligence has advanced molecular structure generation but designing selective inhibitors against protein homologs remains difficult.

Purpose of the Study:

• To develop a de novo structure generator using reinforcement learning for simultaneous multiobjective optimization.
• To create a tool capable of designing selective inhibitors against various protein homologs.

Main Methods:

• Developed a reinforcement learning-based de novo molecular structure generator.
• Applied the generator to optimize 18 objectives, including selectivity and drug-like properties.
• Focused on designing selective inhibitors for tyrosine kinases.

Main Results:

• The structure generator successfully proposed selective inhibitors for tyrosine kinases.
• Achieved simultaneous optimization of inhibitory activities, pharmacokinetic endpoints, and drug-like properties.
• Demonstrated the ability to optimize 18 distinct objectives.

Conclusions:

• The developed structure generator and optimization strategy advance the design of selective inhibitors.
• This approach contributes to the practical application of AI in computational drug design.
• Enables the development of more effective and targeted drug candidates.