FraHMT: A Fragment-Oriented Heterogeneous Graph Molecular Generation Model for Target Proteins

Area of Science:

• Computational Chemistry
• Drug Discovery
• Cheminformatics

Background:

• Molecular generation is crucial for identifying novel drug candidates.
• Existing models often rely on SMILES strings or molecular graphs, limiting feature learning.
• Incorporating substructure information can lead to richer chemical representations.

Purpose of the Study:

• To develop a molecular generation model that leverages substructure information.
• To improve the learning of chemical rules and structure-feature relationships.
• To generate novel molecules with enhanced binding affinity to target proteins.

Main Methods:

• Fragmented molecules into heterogeneous graph representations using atom and fragment data.
• Utilized an encoder-decoder architecture with a self-regressive generative model.
• Applied transfer learning with active ligand molecules for targeted generation.

Main Results:

• The model achieved competitive performance against state-of-the-art methods.
• Generated valid, diverse molecules with desirable physicochemical and drug-like properties.
• Produced novel molecules exhibiting high docking scores against target proteins.

Conclusions:

• Emphasizing substructure information enhances molecular generation capabilities.
• The developed model is effective for discovering novel drug candidates with improved target binding.
• This approach offers a promising direction for computational drug discovery.