REINVENT

REINVENT 2.0: Deep Learning–Based De Novo Molecular Generator

REINVENT 2.0 generates novel chemical compounds using deep learning models that operate on graph-based or string-based representations, including SMILES, to explore and optimize chemical space.

Key Features:

• Deep Learning Architectures: Implements AI models using graph representations and SMILES strings to generate chemical structures.
• Generative Model Framework: Supports generative models such as "reinvent/data/augmented.prior" that output SMILES strings representing candidate compounds.
• Transfer and Reinforcement Learning: Enables focused agents trained via transfer learning or reinforcement learning to target specific regions of chemical space.

Scientific Applications:

• Drug Discovery: Generates and optimizes candidate molecules to support exploration and exploitation of chemical space for therapeutic lead identification.

Methodology:

Employs AI-driven generative models trained on large chemical datasets to produce diverse SMILES-encoded structures. Models iteratively refine compound generation under defined property constraints using transfer learning and reinforcement learning to optimize target-specific outcomes.