Artificial Intelligence (AI) in Proteomics - at ABRF 2025!

Artificial intelligence stuff has come so far in such a short time. Just last year, a sick 3 year old and I could not get either of the ones I pay for to generate an image of a turtle swimming in ketchup. And - today? Check that out! Totally did it! The future is now! 

Okay - so in what might possibly be slightly more useful - what about a killer session at ABRF 2025 on AI in proteomics? This featured two young scientists who actually know what they're talking about. 

Sebastian Paez from Talus bio - who went through where and how we're using AI (surprising places! like inside some of the instruments?? what??) and how our "raw" data has already been manipulated a whole lot. Right now we use our old fashioned search tools and then we go back with AI learning machine things and clean them up. He emphasized new tools where the smart computer is on the front end rather than cleaning stuff up. He also showed a cautionary tale where a modification searched in closed vs open search resulted in completely different results. 

The session ended with Justin Sanders from the Noble lab at University of Washington giving the best description of how Percolator (and derivative tools) work that I've ever seen. Really cool stuff in the context of where these things succeed (how we know) and where they are still not very good (PTMs).

A big surprise in this session was Graham Wiley at the University of Oklahoma. My picture from that was even worse than the ones above. My wife ordered me a new phone this morning. The battery couldn't survive a 5 hour flight home on airplane mode which made finding an uber at 2 in the morning a lot of fun. 

Dr. Wiley is a genomics guy who is doing tons of proteomics now thanks to O-Link technology. Like the really good stuff (5,100 proteins). While the sesssion was AI, as a side effect, I couldn't scribble notes fast enough. 

Like every question that I've had about -

How much does it cost? (If you've already got an Illumina Nova) it'll add about $350k in mandatory robotics. (List price, so probably a lot less). 

What's your throughput? One tech can do about 172 samples in a day (sample prep side) with all the nice robotics. 

What's it cost per sample reagents and labor? His group (which is a certified global service provider) is running about $406/sample in costs. 

What are other limitations to consider? Runs on the Nova thing are very sample specific. Like you can't fire the thing up with plasma in some wells(or equivalent?) and cell lysate in the other. You need those runs to be separate. 

A big advantage for his group has been using the nice robots for other purposes when they aren't running these preps. Again, this seems like I got distracted, but it was a super valuable addition for me for this cool session!  If you're interested in the deep O-link stuff you can find more about his lab and services here.