This new preprint outlines a project that could be termed "ambitious"....
Wooo....ummm.....shit..... I was really going to try and write something clever about this that at least created an illusion that I had some idea what was going on here. Nope. Not going to happen.These people took an absurd amount of data and tried to come up with a better way of predicting protein-protein interactions. We have great actual protein-protein interaction data like BioPlex, so it makes sense that if your fancy artificial netneuro learny machine doohicky could learn from the sets and match the data from BioPlex, you're probably on the right path, since that's kind of the gold standard and everything. I think they show that in one of the curves.
After kind of being able to follow along for the length of this paper, it seems like when you go to the hu.MAP 2.0 database that you'll be blown away by the awesome and powerful data at your fingertips... so I punched in a couple of proteins I know really well....and....
....I suspect it's way cooler if you read the instructions. If "RAS interacts with RAS Interactor #1" is what came out of 15,000 RAW files, we might still be a lot closer to...
....than Cyberdyne System (Boston Dynamics) rolling out a Terminator.
No comments:
Post a Comment