Mass spectrometry based proteomics people will do just about anything to avoid a zero value. We'll just put in random numbers or poorly estimate our signal to noise - and then put in the noise value. There are probably 10 posts over the years on this blog about how to "impute" that zero to a number. And those posts are the ones that will likely have someone trying to figure out how to put a comment on the blog post (which generally doesn't work, but I can't disable that feature on the blog).
Other fields, like single cell seq operate under the assumption that every zero value is actually telling us stuff. Here is my favorite paper on the topic by Stephanie Hicks, who we should totally invite on the podcast....
BIND is a new preprint that I haven't read, but I am currently poking around in their sweet data portal that seems to take LCMS proteomics data in a similar direction.
Preprint link here
As with anything I read that has a lot of formulas and presumably Greek letters (?) after I'm done rolling my eyes and thinking
I look for how they validated whatever those things mean. In this case the authors pull down multiple datasets that I know very well including 2 versions of the NCI-60 cancer panel deep proteomes and one of my very favorite label free single cell proteomics studies. (Jurkat vs HeLa Exploris 480).
And it really looks like this tool kit adds value to there re-analyses. Which I hoped you'd guessed since I took time to write about this preprint.
I do have a gripe that the preprint says the data is formatted into tables in the supplemental. It's not.
But you can run demos on the website with some of the data.
No comments:
Post a Comment