Fig. 9: Evaluation of predicted inter-domain interactions. | Communications Biology

Fig. 9: Evaluation of predicted inter-domain interactions.

From: Multi-domain and complex protein structure prediction using inter-domain interactions from deep learning

Fig. 9: Evaluation of predicted inter-domain interactions.

a Correlation between predicted atomic coordinate deviation (pACD) and true atomic coordinate deviation (tACD) (Pearson r = 0.80, R2 = 0.63, n = 2,190). b Swarm plots displaying the errors (n = 219) of predicted inter-domain distance obtained from different methods. Each point represents one sample with the mean errors marked by a black “|”. The averages are 0.560, 0.724 and 1.151 Å for DeepAssembly, AlphaFold2 and RoseTTAFold, respectively. c Ablation study of inter-domain distance prediction accuracy. “Baseline” denotes the full information is used. “No inter-domain features” denotes a version of DeepAssembly without inter-domain features. “No PAthreader” means that the PAthreader templates are replaced with those searched by HHsearch. “No templates” represents a model without template features. The averages are 0.560, 0.580, 0.655 and 1.621 Å, respectively.

Back to article page