Protein-protein interactions (PPI) are a core dimension in biology as few proteins function in isolation. Recently, we and others have advanced the innovative protein correlation profiling (PCP) method for the large scale analysis of protein-protein interactions. In our PCP analysis we have combined data from mouse liver lysate separations using state-of-the-art chromatographic columns for either size exclusion (SEC), strong anion exchange (SAX), or hydrophobic interaction chromatography (HIC). In addition, we have used comprehensive proteome abundance measurements from both liver lysate subcellular fractionation experiments and cell-type specific proteome datasets.

Our data analysis pipeline uses binary comparison scores between each protein profile within each separation including Pearson correlation, cross-correlation and co-apex scores. These scores are used as features for each binary protein pair in our subsequent random forest machine learning approach. From this analysis we selected 116680 binary interactions with a score greater than 0.5 from the machine learning predictor. These binary pairs were integrated into a non-redundant network using the Clustering with Overlapping Neighbourhood Expansion (ClusterOne) package, while optimising the software settings for best precision and recall. This analysis identified 579 distinct liver protein complexes using stringent settings.

This dataset is quite novel as many of the detected protein complexes contain proteins that are only expressed in liver cell types. These interactions have therefore escaped detection in previous large-scale analyses of protein-protein interactions, which have focused on cultured cancer cell types. Follow-up experiments using immunoprecipitation-MS analysis of individual protein complexes is being used to confirm several of the novel interactions observed in liver tissues. The mouse liver interactome reported here will be extremely valuable for future experiments that examine how protein-protein interactions change after a perturbation such as either metabolic stress, or drug treatment.