I do not 100% get this, maybe, but maybe it will start to fall into place as I write this post about this new study!
If the goal is to understand what is different from healthy to not healthy in something as crazy complex and ultra-high dynamic range as plasma, maybe we need tools that can appropriately discriminate differences.
What I think they did was spike things of known concentrations (yeast digest and E.coli digest) and a learning machine used those known quan markers to….well…learn from…before looking at patient plasma without spikes.
Everything was depleted using a robot (KingFisher, I think, but I’m too lazy to look back at the text now) and it turns out that the machine, once it learned, could find far more interesting markers between healthy and control than non-machiney and learnedy approaches could.
Okay, maybe I do get it. Or maybe I missed the point entirely. Everything was done with DIA and SpectroNaut and the machine got all smart using a long string of different R packages. Honestly, as much as I’m giving R a hard time this week because I’m forced to have it open, for the number of steps this group used, it’s surprisingly easy to follow. Now, will R know that you don’t have some dependency installed that….obviously….you should know you needed…seriously, what is wrong with you? It will, because, of course it will, but will it help you find said dependency? Obviously it is my fault for not having it installed, but this approach might be powerful enough to suck it up and just get typing!
