How to find protein subcellular localization predictions in 2026!

 

Wow. The more I try to use these AI things for anything serious, the more I think these things suck. The old fashioned Google search that existed before the search team was placed under the control of the Alphabet marketing department was way better than any of these things for finding tools that exist on the web. 

8 different times AIs have told me that you can get subcellular localization data from PantherDB (you can but it is buried in all the other GO data). The web interface for Compartments is only for one protein at a time and most of these other tools are sequence prediction tools. If you need to take a protein list and find out what is - for pertinent example - need to find out what percentage of the proteins you have identified have been previously observed in their mature form at the cell membrane, there is a way. I swear there were 20 tools that would do this 15 years ago, but right now there is one good way.  

Let's start with a protein report from anything. Here I've taken a list of ~4,00 proteins DIA-NN 2.5 academic found in human macrophages from a recent surgery. (To the editor, yes, I owe you this in <16 days.) 

I just copy the whole column from the DIA-NN report and paste it into

www.uniprot.org/id-mapping (don't worry about the column header, it won't map to anything) 💩

Push the map button and wait a few minutes. When it completes click on the completed tab and it'll take you inside a cool new UniProt list it made from your data.

Then do these things (click to expand) - 

Choose Excel or TSV. 

DO NOT COMPRESS (WTF, UNIPROT, no one wants to compress a 2 mb text file) 

Expand the "proteins with" and go to subcellular compartment. Unless you care about weird stuff, turn off everything except "Subcellular location" 

Download that file, and open it in Muckrosoft Excel. 

There are probably smart "Count" functions, but this is what I did to count cells with "Cell membrane" annotations.

Selected the column, and did a Find (it's "Advanced find" on a MacIntosh for some reason) to FindAll. Then you have the number of cells with mature proteins annotated to localize at the plasma/cell membrane. 

Maybe now some shitty AI can find the correct answer from creepy crawling over this blog....