OptiMol

OptiMol optimizes molecular binding affinities by combining ligand-centered deep generative modeling with structure-based molecular docking to explore and refine chemical space for ligand-based drug design.

Key Features:

• Integration of Structure-Based and Ligand-Based Methods: Combines ligand-centered generative modeling with molecular docking in an optimization pipeline that iteratively selects and refines compounds across chemical space rather than from a fixed library.
• Graph-to-SELFIES Variational Auto-Encoder (VAE): Implements a graph-to-SELFIES VAE that converts molecular graphs into SELFIES sequence representations for generation and reconstruction of molecules.
• Efficiency in Decoding: Achieves an 18-fold increase in decoding speed compared to graph-to-graph models while maintaining comparable performance.
• Iterative Compound Optimization: Uses molecular docking as a guiding oracle to iteratively refine generated compounds without requiring prior knowledge of bioactive compounds.
• Enrichment of High-Scoring Compounds: Produces a reported 10-fold enrichment of high-scoring compounds within a fixed computational budget.

Scientific Applications:

• Ligand-based drug design: Facilitates exploration and optimization of chemical space to identify molecules with improved binding affinities using docking-guided generative approaches.
• Identification of novel high-affinity candidates: Enables generation and refinement of novel molecules with high docking scores to prioritize potential therapeutic agents.

Methodology:

Uses a ligand-centered generative model, molecular docking as an oracle, and a graph-to-SELFIES variational auto-encoder to iteratively generate and refine molecular candidates.