Deep reinforcement learning in chemistry: A review.

Area of Science:

• Computational Chemistry
• Artificial Intelligence

Background:

• Reinforcement learning (RL) demonstrates significant success across diverse computational chemistry domains.
• Key application areas include molecule generation, geometry optimization, and retrosynthetic pathway searches.

Purpose of the Study:

• To motivate the application of RL in chemistry.
• To provide a foundational understanding of RL formalism for chemistry problems.
• To review recent advancements and inspire novel RL applications.

Main Methods:

• Discussion of RL formalism relevant to chemistry.
• Analysis of solution formulations and algorithms from recent literature.
• Comparison of different approaches and their underlying RL algorithms.

Main Results:

• Overview of the current state of RL applications in chemistry.
• Identification of actively researched open problems.
• Insights into utilizing RL for chemical challenges.

Conclusions:

• RL is a powerful tool for advancing computational chemistry.
• Understanding RL principles enables chemists to tackle complex problems.
• The review aims to foster innovation in RL-driven chemical research.