SE(3)-equivariant ternary complex prediction towards target protein degradation

Area of Science:

• Biochemistry and Structural Biology
• Computational Chemistry
• Drug Discovery

Background:

• Targeted protein degradation (TPD) utilizes small molecules like PROTACs and molecular glue degraders (MGDs) to eliminate disease-causing proteins.
• Rational design of TPDs is challenging due to the difficulty in determining the structures of ternary complexes formed between target proteins, E3 ligases, and degraders.

Purpose of the Study:

• To develop a novel deep learning approach, DeepTernary, for accurate and rapid prediction of protein-ligand-E3 ligase ternary complex structures.
• To assess DeepTernary's performance on both PROTAC and MGD benchmarks and its correlation with experimental degradation potency.

Main Methods:

• An end-to-end deep learning model, DeepTernary, employing an SE(3)-equivariant encoder and a query-based decoder.
• Training on the curated TernaryDB dataset.
• Evaluation using PROTAC and MGD benchmarks with a blind docking protocol.

Main Results:

• DeepTernary achieved state-of-the-art performance on PROTAC benchmarks without prior exposure.
• Demonstrated notable prediction capability on the more challenging MGD benchmark.
• Predicted structures showed correlation between buried surface areas and experimental degradation potency.

Conclusions:

• DeepTernary provides a powerful computational tool for accelerating the rational design and development of targeted protein degraders.
• The method advances the field of structure-based drug design for previously undruggable targets.