Figure 3.

Expected accuracy based on the percentage of sequence identity between targets and templates. 38 463 predicted protein complex models were used to derive expected (A) structure and (B) interface accuracy. The structure confidence score is calculated by mapping each model's identity score to the likelihood that the model is less than 10 Å all-atom RMSD to the native structure. Likewise, the interface confident score is calculated by mapping the identity score to the likelihood that more than 50% of the interacting residues are correct.