Over 9,000 LC-MS/MS experiments integrated by machine learning!

This is AWESOME! Before I get too carried away, let's point you to the very nice open access paper here!...then...

...put it into some context!

There have been multiple huge attempts to manually map protein-protein interactions. I have been completely unfiltered in my love and respect for the BioPlex project, and this will not change, but there are other resources with other technologies as well. BioPlex is a reasonably new effort and I've tried not to seem too stressed out about new studies where people have met-analyzed other datasets, like the huge Y2H (yeast 2 hybrid) assays when BioPlex info is available.

TADAA!!!  Welcome to hu.MAP (proteincomplexes.org)

What is it? It is a meta-analysis of the BioPlex data released so far with:
this study from Marco Hein et al., (great Max Planck study in Cell from 2015) and
this cool one from Cuihong Wan et al., that appeared in Nature around that same time (that I missed till now)

These studies are all great and well done and just awesome on their own. Why would anyone mess with them? Answer: Because that is what public repositories of data are for. And -- get this -- analyzing these together with fancy machine learning algorithms -- turns up protein-protein interactions that none of them did on their own!

Not to leave it alone there -- no way -- these authors also look at some of the Y2H databases, which is cool and all -- but they painstakingly validate some of the protein-protein interactions that their methods pull out of these beautiful data sets and show they are VERY VERY real.

How'd they do it? Magic formulas with loads of Greek letters that I can, in no way, confirm are correct or accurate -- but their validation assays sure look great! The important part is that we can go to proteincomplexes.org and simply type in what we're interested in and yield the rewards of their efforts!