A method for predicting protein-protein interaction types

Yael Silberberg, Martin Kupiec, Roded Sharan

Research output: Contribution to journalArticlepeer-review


Protein-protein interactions (PPIs) govern basic cellular processes through signal transduction and complex formation. The diversity of those processes gives rise to a remarkable diversity of interactions types, ranging from transient phosphorylation interactions to stable covalent bonding. Despite our increasing knowledge on PPIs in humans and other species, their types remain relatively unexplored and few annotations of types exist in public databases. Here, we propose the first method for systematic prediction of PPI type based solely on the techniques by which the interaction was detected. We show that different detection methods are better suited for detecting specific types. We apply our method to ten interaction types on a large scale human PPI dataset. We evaluate the performance of the method using both internal cross validation and external data sources. In cross validation, we obtain an area under receiver operating characteristic (ROC) curve ranging from 0.65 to 0.97 with an average of 0.84 across the predicted types. Comparing the predicted interaction types to external data sources, we obtained significant agreements for phosphorylation and ubiquitination interactions, with hypergeometric p-value = 2.3e -54 and 5.6e-28 respectively. We examine the biological relevance of our predictions using known signaling pathways and chart the abundance of interaction types in cell processes. Finally, we investigate the cross-relations between different interaction types within the network and characterize the discovered patterns, or motifs. We expect the resulting annotated network to facilitate the reconstruction of process-specific subnetworks and assist in predicting protein function or interaction.

Original languageEnglish
Article numbere90904
JournalPLoS ONE
Issue number3
StatePublished - 13 Mar 2014

All Science Journal Classification (ASJC) codes

  • General Biochemistry,Genetics and Molecular Biology
  • General Agricultural and Biological Sciences


Dive into the research topics of 'A method for predicting protein-protein interaction types'. Together they form a unique fingerprint.

Cite this