Collective intelligence defines biological functions in Wikipedia as communities in the hidden protein connection network

Andrei Yu. Zinovyev, Urszula Czerwinska, Laura Cantini, Emmanuel Barillot, Klaus M. Frahm, Dima L. Shepelyansky

Abstract: Author summary The long-standing effort for annotating protein functions from published experimental evidences is still far from being completed, partly due to a limited number of biocurators involved in it. Wikipedia was thought to be a suitable platform for the protein function curation crowdsourcing through exploiting the wisdom of the crowd principle. Starting from 2008, English Wikipedia was automatically populated with thousands of protein pages and links between them (Gene Wiki project), which created a useful and rapidly evolving knowledge resource. However, it remains unclear what is the benefit of hyperlinking protein pages with the whole Wikipedia knowledge corpus. We applied the recently introduced network analysis method, called reduced Google Matrix (REGOMAX), in order to study the structure of direct and indirect (hidden) links between protein pages through the rest of the global Wikipedia network. As expected, the network of direct links had node degree distribution approximately following the power law. In contrast, the network of hidden links was characterized by larger than expected tight communities of proteins related to their known functions, such as involvement in immune system. The “friendship network” of these protein groups can be used for automated annotations of their functions from non-protein Wikipedia pages. We estimated the size of the expert Wikipedia contributor community, specifically working on protein and associated pages, to be nearly 1000 wikipedians with primarily biomedical background. We conclude that the structure of global Wikipedia network can improve the annotation of protein functions by amplifying the wisdom of the crowd effect.

External IDs:dblp:journals/ploscb/ZinovyevCCBFS20