bioengineer | @try_litefold

conserved motif
Joined April 2025
it's a privilege to do hard things
2
1
15
351
did some simulations on a wide range of targets a while back and this pretty much holds true
Replying to @RolandDunbrack
My current mental model of ipSAE is that the low end (depends on the target) is extremely predictive of nonbinding, but higher values are not. So it is useful but not perfect at filtering. Per @amirshanehsaz's comment, 11/60 ain't bad!
1
5
478
again, simulations doesn't mean it binds but i was assessing complex stability
51
nabajit (bio/acc) retweeted
As said before, if we are going to do agentic science, lets do it in a right, controlled way, just like how controlled reproducible experiments looks like. All u gotta do is just connect ur lab.
Ahem ahem. Topic: Own and Customize everything 294 Connectors across target science, mol engineering, developmental biology. DBs, tools, atlases, just ask for it. You can add custom ones. 200+ skills along the same lines. Add custom skills. Give permissions to all the platform workflows. Add custom instructions, give permissions what your agents can and can not use and when to ping you. Even create your own subagents. Compute: Select the compute capacity you need when agentic coding requires gpus. We have found ways where we have ontologically organized every connector and skill, use jev-like decision models, and cut the token cost by half, and actually getting better accuracy in using the right things in the right time. As said before, if we are going to do agentic science, lets do it in a right, controlled way, just like how controlled reproducible experiments looks like.
2
3
11
618
auto translated chinese tweets are banger
1
4
297
man i miss lucknow
mf will buy an iphone n then suddenly act like a photographer i am mf
1
14
411
nabajit (bio/acc) retweeted
Glycosylation is among the most important modifications in biology. We know relatively little about glycobiology, because it is hard to study. This Nat Rev Methods primer describes our best tools for analyzing glycosylation 🔽
5
65
3
376
16,496
Sovereign foundation models are the only way forward. The sooner we do it the better we are.
1
7
468
this is so cool
how do the 100 blockbuster drugs actually work? I mapped every one onto its 3D target structure and explained how they work. Small molecules to ADCs. atlas.dominikscherm.de
11
1,147
nabajit (bio/acc) retweeted
Replying to @lydiahallie
wait is it just me or is Opus 5.5 dumber
549
408
352
10,901
944,558
nabajit (bio/acc) retweeted
Aside browser is really good. Every website feels like an app.
5
17
3
207
21,168
With open models like RFD4, I wonder how far ahead closed-source models like Chai Discovery are.
6
1
1
41
4,175
Hill climbing biology is coming
I asked Google CEO Pichai what happened to the missing Gemini 3.5 Pro model that never shipped. He gave me an interesting window into the hard decisions big labs have to make in model development: "While training the 3.5 series, we had Flash and Pro, which we were working on. An area we internalized through the middle of this year was that there were gaps in agentic capabilities, particularly around coding... We wanted to definitely hill climb there, and the team felt the fastest way to do so was using Flash. It's a smaller model and more efficient, so you can really get a lot more iterative loops there. That's what led to 3.5, 3.6, 3.7, and 3.8 Flash. By the time we got that done, we had a chance to apply it to Pro, but Argon pre-training was done."
1
6
566
life in past few weeks
4
40
985
Strava doesn't expose MCP without a premium subscription 🫠
1
293
nabajit (bio/acc) retweeted
PhD teaches you time is cheap and money is expensive. Startup teaches you the inverse.
55
211
17
3,344
91,824
RFD4-Proteina's documentations are really underrated
1
3
57
2,936
RFD4 Proteina is super fast. They optimized it so well
1
1
25
1,624