Wow. I'm completely blown away by this one. If you've got VERY GOOD AD BLOCKERS INSTALLED, you can find the popsci article here. Don't click it if you don't. Hot dog, the little counter on the corner of my screen rang up 74 ads and trackers in just a second. And the banner at the top warned me of "consequences" of blocking those 74 ads.
The article is sort of about this new PNAS paper! Click this instead. My school doesn't subscribe to PNAS? Weird. Maybe my ad blockers are too good.
Okay, but here is the thing. What if you could convert protein sequences to text? For real, hear me out here. You could, in theory, assign each amino acid IN ORDER an alphabetical letter. Then you could take all the complexity of a protein sequences - even one with hundreds of amino acids! - and make it just a short paragraph!
See where I'm going with this? I bet some of you super saavy scientists do. Good for you! Then you could use that text input into - you guessed it - these amazing LLM things!
I suspect this group at the NIH did a far better job than me when I just tried it with the LLMs I have access to, but I swear I think I just found an interesting pattern in my proteomics data with a similar approach. I have this output I haven't ever seen before.
Serine-Histidine-(Leucine, or Isoleucine, but likely the latter)-Threonine-Threonine-Tyrosine-Serine-Histidine-(Leucine, or Isoleucine, but likely the latter)-Threonine-Threonine-Tyrosine-Serine-Histidine-(Leucine, or Isoleucine, but likely the latter)-Threonine-Serine-Histidine-(Leucine, or Isoleucine, but likely the latter)-Threonine
Have I discovered an entirely new and obviously important protein motif? It seems worth investigating for potential phosphorylation sites alone!
No comments:
Post a Comment