I did this at home with a 5090 on a different target a few weeks ago. Cannot generate 318,000 designs but Claude chose the nearly same tooling, except I used BindCraft2 which is better for multi target designs in the beginning. This isn't that special, its just doing tool calls. I got 100 or so trajectories running 24hrs just for fun.
"Throughout the process, Claude generated 318,484 protein designs and selected 180 of them for testing. Our team added 31 more, with Claude’s help."
"lab-tested all 211 designs to see if they would bind to the toxins they were designed against, and 17 designs met our criteria for a binder. "
Alright how many of those 17 were 100% Claude, I know mine were. These mini-proteins the tooling I'm told is pretty good for these. I'm worried that Claude might just choose the same tooling for everyone if this is the path we go down and designs stagnate?
There’s certainly value in using LLM to search unknown spaces for research of knowledge captured in human text forms, but having done a lot of modeling and simulation in applied sciences over my career, I often wonder for some of these cases how much is just having people interested in various computational sciences, modeling and simulation, and a massive pool of compute to deploy things and approaches to rapidly wire things together on the cheap, sort of like this case.
It was pretty grand not having to string together the bio-informatics and protein design pipeline and let Claude handle that and the environment. A lot of these tools in the past are fragmented and sometimes frozen to certain versions of a language which meant debugging. I didn't have to do at all; if there was env issues or GPU incompatibility Claude ran tests before execution and figured it all out for me. Truly the most boring stuff of this it fixed which would probably have made me give up if I was on my own.