Transformer Decoder Learns from a Pretrained Protein Language Model to Generate Ligands with High Affinity.

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Molecular modeling

Background:

• Drug discovery is a lengthy and expensive process.
• Deep learning offers potential to accelerate the identification of drug candidates.
• Identifying molecules that bind specific protein targets is crucial for drug development.

Purpose of the Study:

• To present a novel deep learning generative model, Prot2Drug, for designing drug-like molecules.
• To leverage pretrained protein language models and transformer capabilities for target-specific ligand generation.
• To accelerate the drug discovery pipeline by suggesting high-affinity ligands for proteins of interest.

Main Methods:

• Developed Prot2Drug, a deep learning generative model.
• Utilized pretrained protein language models for target information.
• Employed transformers to learn from extensive protein-ligand interaction data.
• Generated novel compounds with predicted favorable physicochemical properties and target affinity.

Main Results:

• Prot2Drug successfully generated novel ligands with high predicted affinity for specific protein targets.
• The model reproduced known protein-ligand interactions.
• Identified potential drug repurposing candidates by suggesting new targets for known compounds.
• Demonstrated efficacy even for protein targets with limited structural or ligand information.

Conclusions:

• Prot2Drug significantly accelerates drug discovery by generating targeted ligands.
• The model's ability to learn from diverse data enables effective ligand design for challenging targets.
• Prot2Drug shows promise for identifying novel therapeutics and drug repurposing opportunities.