Fig. 4: Comparative performance and feature analysis for protein-protein structural similarity prediction (n = 2701 pairs).
From: High-accuracy protein complex structure modeling based on sequence-derived structure complementarity
a Performance comparison between our method and PLMsearch based on Pearson correlation, Spearman correlation, and ROC AUC metrics. The data are presented as the mean, with n = 2701 pairs from the benchmark set. b ROC curves showing the classification performance of our method, ESM2, and Sequence representation, with n = 2701 pairs from the benchmark set. c Correlation analysis between ESM2 embeddings and additional sequence-derived features, with n = 2701 pairs from the benchmark set. d Feature distributions and pairwise correlations, with n = 2701 pairs from the benchmark set. Histograms represent the distribution of each feature, scatter plots show pairwise relationships, and upper triangle statistics include Pearson, Spearman correlations, and AUC values. Source data are provided as a Source Data file.
