#ai #llm #bioinformatics
Surprise! AI might be useful for science after all.
A bioinformaticist friend writes:
Well, looks like I've been replaced by AI.
Here is a paper from Aviv Regev's group that uses multi agent models (LLMs that talk to each other) to basically do one of the pipelines that my group was responsible for at ReadCoor.
One task that I am happy to see that they included was building probe panels, which are sets of (usually) genes that can be used to see what is going on inside a sample. Five years ago, this was a very onerous task that included loads of manual literature search and curation. I can't imagine doing it now without at least starting with a Deep Research or similar LLM query. (I was the manager so my role in this was to repeatedly say "Wow. That looks awful. Thanks for doing that.")
There is stuff around the AI that is I think is pure hype, but this is the kind of practical, important task that most humans hate and that AI is really great at.
Pretty neat stuff!
*It did take them 30 more people to build the AI than we had to build the pipeline. But that'll go down.
**I should add that our panels were really great. During AGBT 2020, people who came to our booth were gathered around the screens showing our panels. Seeing the genes they cared about included really gave people an understanding of the potential for in situ sequencing.
https://www.biorxiv.org/content/10.1101/2025.04.03.646459v1