Network-based prediction of protein interactions

Abstract

Despite exceptional experimental efforts to map out the human interactome, the continued data incompleteness limits our ability to understand the molecular roots of human disease. Computational tools offer a promising alternative, helping identify biologically significant, yet unmapped protein-protein interactions (PPIs). While link prediction methods connect proteins on the basis of biological or network-based similarity, interacting proteins are not necessarily similar and similar proteins do not necessarily interact. Here, we offer structural and evolutionary evidence that proteins interact not if they are similar to each other, but if one of them is similar to the other’s partners. This approach, that mathematically relies on network paths of length three (L3), significantly outperforms all existing link prediction methods. Given its high accuracy, we show that L3 can offer mechanistic insights into disease mechanisms and can complement future experimental efforts to complete the human interactome.

A reference map of the human protein interactome

“…However, proteins with a tendency to share interaction partners often have interaction interfaces that are similar, as opposed to complementary, and therefore tend not to interact, unless both proteins originate from the same ancestral protein that was able to self-interact 27 (Extended Data Fig. 4c).…”

“…Finding a new protein annotation can be described as a link prediction problem between a node representing the function and the proteins. Initially, we connect the functional node to each of the proteins annotated with this function and obtain a link prediction score for each other gene in the network based on our recently developed link prediction method 27 . As the result, the indirect score of protein i is obtained as where aik is the connection weight between nodes i and k and kj is the degree of node j.…”

Structure-based validation can drastically under-estimate error rate in proteome-wide cross-linking mass spectrometry studies

“…have leveraged known protein-protein interaction networks to visualize the XL-MS datasets and infer biological insights 14,15 . Using a metric analogous to a common machine learning term, precision 16,17 , we leverage prior knowledge of experimentally-detected protein interactions to calculate the 'fraction of interprotein cross-links from known interactions' (FKI) and provide a comparative quality metric (Methods).…”

Visualizing Human Protein‐Protein Interactions and Subcellular Localizations on Cell Images Through CellMap

“…Experimental evidence for protein subcellular localization [also referred to as “cellular component” (The Gene Ontology Consortium, 2019) or “location”] and protein‐protein interactions (PPIs) is steadily increasing in big protein databases, e.g., UniProt (The UniProt Consortium, 2019). Even for proteins still lacking reliable experimental annotations, many prediction tools exist (Almagro Armenteros, Sønderby, Sønderby, Nielsen, & Winther, 2017; Cong, Anishchenko, Ovchinnikov, & Baker, 2019; Kovács et al., 2019). With increasing availability and abundance of these data, interactive and intuitive ways of visualization on a system level—i.e., for larger data sets rather than individual proteins—become increasingly important, both for research and educational purposes.…”

“…1 of 10 (The UniProt Consortium, 2019). Even for proteins still lacking reliable experimental annotations, many prediction tools exist (Almagro Armenteros, Sønderby, Sønderby, Nielsen, & Winther, 2017;Cong, Anishchenko, Ovchinnikov, & Baker, 2019;Kovács et al, 2019). With increasing availability and abundance of these data, interactive and intuitive ways of visualization on a system level-i.e., for larger data sets rather than individual proteins-become increasingly important, both for research and educational purposes.…”