DeepDegradome: A structure-aware deep learning framework for PROTAC and ligand generation against protein targets

Area of Science:

• Medicinal Chemistry
• Artificial Intelligence
• Drug Discovery

Background:

• Targeted protein degradation using Proteolysis-Targeting Chimeras (PROTACs) is a key strategy in drug discovery.
• Designing effective PROTACs is challenging, particularly for proteins lacking well-defined binding sites, and current methods are limited by fixed ligands and linkers.

Purpose of the Study:

• To introduce DeepDegradome, an AI-powered platform for automated, structure-aware design of small-molecule ligands and PROTACs.
• To overcome limitations of current PROTAC design methods by enabling de novo construction of ligands and PROTACs.

Main Methods:

• Utilizing a large fragment library from public databases and an in-house docking method (iFitDock) for initial fragment identification.
• Assembling fragments based on target protein pocket features to build novel ligands.
• Constructing PROTACs from generated ligands, independent of predefined warheads or E3 ligases.

Main Results:

• DeepDegradome generated more valid, drug-like molecules with higher predicted binding affinity compared to other AI models.
• Successfully designed and validated potent inhibitors and PROTACs for WDR5 and CDK9 targets.
• Synthesized compound showed high agreement between predicted and experimentally determined binding conformation via X-ray crystallography.

Conclusions:

• DeepDegradome offers a scalable and reliable AI-driven solution for designing both ligands and PROTACs.
• The platform facilitates the discovery of new drugs by enabling innovative molecular designs for challenging protein targets.
• Automated design of PROTACs and ligands accelerates the drug discovery pipeline.