DiffPROTACs is a deep learning-based generator for proteolysis targeting chimeras.

Area of Science:

• Biotechnology and Drug Discovery
• Artificial Intelligence in Chemistry
• Molecular Modeling

Background:

• PROteolysis TArgeting Chimeras (PROTACs) offer a novel therapeutic modality, but rational linker design is challenging.
• Fragment-based drug design (FBDD) is a viable approach for PROTAC development.
• AI, particularly diffusion models and Transformers, shows great promise for molecular generation.

Purpose of the Study:

• To introduce DiffPROTACs, a novel diffusion model for generating PROTAC linkers.
• To leverage Transformers and graph neural networks for enhanced molecular generation.
• To provide a valuable resource for PROTAC research and development.

Main Methods:

• Developed DiffPROTACs, a diffusion model integrating Transformers and O(3) equivariant graph neural networks.
• Employed Transformers for node updates and GNNs for atomic coordinate refinement.
• Trained and validated the model on traditional FBDD datasets (ZINC, GEOM) and a custom PROTACs dataset.

Main Results:

• DiffPROTACs demonstrates competitive performance against existing models on FBDD datasets.
• Achieved a high validity rate of 93.86% for generated PROTACs after fine-tuning on a specific PROTACs dataset.
• Generated a comprehensive database of novel PROTACs for future research.

Conclusions:

• DiffPROTACs represents a significant advancement in AI-driven PROTAC linker design.
• The model's ability to generate valid PROTACs with high accuracy facilitates drug discovery.
• The provided database and code enable further exploration and application of this technology.