Fig. 2: The benchmarking results of PLM-interact compared with state-of-the-art protein-protein interaction (PPI) prediction models: PLM-interact achieves the highest PPI prediction performance.
From: PLM-interact: extending protein language models to predict protein-protein interactions
a The data size of training, validation and test protein pairs. b The taxonomic tree of the training and test species is aligned with the precision-recall curve of each model on each test species. A bar plot of AUPR values illustrates the PPI prediction benchmark. The distribution of predicted interaction probabilities of positive and negative protein pairs for each PPI model is shown in Supplementary Fig. 3. All species icons in panel (b) are created in BioRender (Liu, D., 2025; https://BioRender.com/1ezsj7q). Source data are provided as a Source Data file.
