Fig. 4: DeepTernary achieves the highest accuracy in PROTACs-induced ternary structure prediction.

DeepTernary outperforms existing methods in predicting 22 existing PROTAC-induced ternary structures regarding the metrics of the DockQ score (a) and the first acceptable rank containing at least one prediction with DockQ ≥0.23 (b). For those that failed to generate an acceptable result, we manually set the rank value to 41 for a fair comparison. Besides, DeepTernary achieves better DockQ performance on most complexes compared to the current best model, the RosettaDock-Based model (c). d, e DeepTernary outperforms existing methods on the percentage of High/Medium/Acceptable. f, g, DeepTernary has a higher potential to generate decent (RMSD <10 Å) results. h DockQ performance comparison among different E3 ligases. i Surface illustration of the predicted structure of PDB ID 5T35. j Three examples of predicted ternary structures (teal and orange for the protein and ligand, respectively) overlaid with the experimental structures (gray and green for the protein and ligand). The receptor protein is colored red, and the chemical structure diagrams of PROTAC molecules are illustrated at the bottom. (All box plots in this figure extend from the first quartile (Q1) to the third quartile (Q3) of the data, with a line at the median. The whiskers extend from the box to the farthest data point lying within 1.5x the inter-quartile range (IQR) from the box. Flier points are those past the end of the whiskers. Statistical analyses were performed using the two-sided independent t-test. The sample number (n) is annotated under each plot. Source data are provided as a Source Data file.