Reinforcement learning-based generative artificial intelligence for novel pesticide design.

Area of Science:

• Agricultural Science
• Computational Chemistry
• Artificial Intelligence

Background:

• Pesticides are crucial for food security, driving demand for sustainable

Purpose of the Study:

• To pioneer generative artificial intelligence (AI) for pesticide design.
• To develop a reinforcement learning (RL)-based framework for creating pesticide-like molecules with high binding affinity.

Main Methods:

• PestiGen framework with two components: PestiGen-G (character-based generative model + REINFORCE algorithm) and PestiGen-S (fragment-based generative model + Monte Carlo Tree Search).
• Active learning strategy to minimize false positives.

Main Results:

• Generated molecules exhibit superior pesticide-likeness and binding affinity compared to existing methods.
• Successfully designed a novel 4-hydroxyphenylpyruvate dioxygenase inhibitor (YH23768) with significant herbicidal potency.

Conclusions:

• PestiGen serves as a valuable proof-of-concept tool for pesticide design.
• The PestiGen web server is publicly accessible for research use.