DiffTernary: a Diffusion Model for Molecular Glue Ternary Complex Docking

DiffTernary, trained on Ternary-PDB, predicts molecular glue ternary complexes with acceptable PPI DockQ (>0.23) for 50.0% of MGBench complexes using a single prediction.

• Why it matters: Modeling molecular glue ternary complexes is difficult, and existing docking and structure predictors need better accuracy and reliable binding-pose and relative-arrangement predictions.
• What they did: The method is a diffusion-based docking framework for joint prediction of protein arrangement and MG binding pose, trained on Ternary-PDB (76,702 PDB-derived ternary complexes).
• The result: In head-to-head comparisons, DiffTernary ranked second with a best-of-25 PPI success rate of 78.4%, and a TernaryDB-trained model reached mean DockQ 0.306 (single prediction) on 94 test complexes.

AI summary of the abstract · may contain errors · not medical advice · read the paper.