PETrans: De Novo Drug Design with Protein-Specific Encoding Based on Transfer Learning.

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• Deep generative models have advanced novel drug molecule design.
• Existing methods generate drug-like molecules needing further optimization.
• Limited datasets pose challenges for developing target-specific ligands.

Purpose of the Study:

• To propose a deep learning method for generating target-specific ligands, especially for limited datasets.
• To leverage generative pretraining and protein encoding for guided molecule generation.
• To enhance ligand binding affinity through transfer learning.

Main Methods:

• Utilized generative pretraining (GPT) for molecular contextual feature extraction.
• Employed three protein-encoding methods to capture target protein properties.
• Integrated molecular sequence and protein information to guide ligand generation.
• Applied transfer learning to fine-tune the model for improved binding.

Main Results:

• The model successfully generated novel molecules with higher docking scores for three different targets.
• Demonstrated capability in producing target-specific ligands.
• Validated the effectiveness of combining protein and molecular data for generation.

Conclusions:

• The proposed deep learning method effectively generates target-specific ligands with enhanced binding affinity.
• This approach offers a viable solution for drug discovery with limited target-specific data.
• The model shows promise for accelerating the development of optimized drug candidates.