Conformer-RL: A deep reinforcement learning library for conformer generation.

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Drug Discovery

Background:

• Generating accurate molecular conformations is crucial for understanding chemical properties and biological activity.
• Traditional methods can be computationally expensive and may not capture the full conformational landscape.

Purpose of the Study:

• To introduce Conformer-RL, an open-source Python package for generating diverse low-energy molecular conformations.
• To provide a flexible platform for applying deep reinforcement learning to conformer generation.

Main Methods:

• Utilizes deep reinforcement learning (RL) algorithms.
• Employs graph neural network architectures optimized for molecular structures.
• Offers modular interfaces for custom RL environments and agents.

Main Results:

• Successfully generates diverse sets of low-energy conformations for various molecules, including drug-like compounds.
• Provides tools for visualization and saving of generated conformers.
• Demonstrates applicability to both small molecules and polymers.

Conclusions:

• Conformer-RL offers an efficient and versatile approach to molecular conformer generation.
• The package serves as a valuable resource for both practical applications and research into novel RL algorithms for chemistry.