
Genome engineer, data scientist, garage bio @countrcultrlabs @Regeneron @Columbia
Bay Area
Joined January 2014
- Tweets1.5K
- Following2.1K
- Followers825
- Likes8.2K
Dream collab. I’d buy merch
1/ Today we’re announcing @NOETIK_ai collaboration with @MSKCancerCenter to advance cancer biology world models.
This collaboration brings together MSK’s world class oncology expertise and research resources with NOETIK’s frontier AI and data foundry capabilities.
1
13
683
Jacob Boysen retweeted
Replying to @iskander
The cxg version of the dataset also has an entertaining normalization that zeroes out MALAT1 in 2/3 of the nuclei
Hustling so hard they make Young Jeezy blush
18 paid partnerships on the tweet
Last seen nuking a 200mn bio startup in 18 months
YC is involved
Literally the Theranos pitch
🤷🏻♂️ 🤷🏻♂️ get out there and get some anon 🤷🏻♂️ 🤷🏻♂️
Introducing Helios: measuring 1,000+ protein biomarkers from a drop of blood to help you live younger, longer.
Here’s how it works:
• Self-collect blood at home with a painless, needle-free device
• We measure the proteins carrying signals from cells and organs throughout your body
• Your Helios AI agent explains your results, highlights risks, and offers personalized guidance
Proudly backed by Y Combinator and Formation VC.
Join the waitlist at myhelios.com
Jacob Boysen retweeted
But this crap is such a constant issue with Anthropic employes. Amanda Askell does this, too. On the one hand, they want to be the main character -- figures of world-historic importance. Players on the world stage. But when attacked, it's "I'm jus a smol bean why u so mean?"
All of this data is entirely consistent with a scenario where evals team found traction in curating protein scoring rubrics (easy to imagine) which anthro threw compute at and hill climbed. Does not imply extravagant ‘emergent bio’ reasoning, just questionable capital spend
Can an LLM fold a protein from sequence alone?
@sokrypton's contact-map experiment is a neat idea, but the test is toy-sized: [small, well-studied protein — check against his post].
We reran the same protocol on a PDB entry released in Oct 2026. The 3D result is near random. 🧵
Jacob Boysen retweeted
Replying to @sokrypton @proteinrosh
I guess another solution is just to ask it to return a contact map. Here is Fable output:
This stuff almost always comes down to filtering—either computationally or at the literal vacuum hose. Lots of butterfly wings to disturb
“Wet-lab validated.” AI binder design papers keep reporting impressive hit rates. But does it bind in a second lab?
One independent retest: only 29/107 published AI-designed nanobodies showed binding—and 6 of those had high aggregation.
Germinal stands out for the wrong reason: just 1 of 12 published binders showed detectable binding on retest—a result that warrants an explanation.
How much trust should we put in headline hit rates?
Read more on my blog 👇
smiles724.github.io/blog/ai-…
11
1,309
Jacob Boysen retweeted
Why one standard matters: these datasets overlap each other far more than people assume.
88% of ProteinGym proteins have a homolog in a dataset measuring a different function. The same holds for 88% of AlloBench and 89% of BindingDB. CAFA5 and PPI overlap at 80%.
We @Dyno_Tx (+ our partners) can do that with 2% of that budget, probably faster than them if they set out to do it tomorrow.
Not to say I'm not excited for what they will do with this allocation. Their work is pioneering and inspiring.
But curing 1000s of monogenic disease is likely too hard for them in the next few years:) It's not just an AI problem. Requires very high integration between AI and lab/clinic to make the AI count.
Pretty obvious why 5.5 is good at predicting mammalian coding proteins but not non-coding eg promoters—one is a hell of a lot easier to pattern match than the other
Replying to @TuXinming
DNA is different: Opus 5.5 predicts human and mouse coding sequence, but not the rest of the genome. It learned proteins, not genomes.
Code, data and every model reply: github.com/XinmingTu/LLM-bio…
Full detail and more results Post: xinmingtu.cn/blog/2026/claud…
Jacob Boysen retweeted
Cell culture decisions depend on years of scientific expertise. Today we're launching a new product page showing how our cloud-based microscopy and AI software helps scientists:
🔬 Centralize and review microscopy images across instruments, experiments, and timepoints.
✨ Accelerate analysis and decision-making with AI models that learn from scientists' feedback.
🤝 Capture and share scientific expertise to make cell culture workflows more consistent, collaborative, and reproducible.
Bring a plate you've already imaged. We'll show it to you in Culture Monitor. Book a demo!
Jacob Boysen retweeted
"Claude was trained on protein sequences at scale" by @TuXinming at Phylo 🧬
Something I think has been suspected before, so it's interesting to see more evidence
If true, I'd be interested to see if/how the open model ecosystem follows suit
Jacob Boysen retweeted
Do you need to build & test 100s to 1000s of protein expression plasmids? Introducing a new design-build-test pipeline that increases throughput & lowers costs by ~24-fold! Stress tested on complex proteins!
pubs.acs.org/asbcd6/article/…
We streamline & accelerate steps combining:
/1
I don’t ✋🏼want 2 👁️any more PRO⛹️TISTS🕺🏻 in UR hype video if U 🫵🏻don’t know who NOMARSKI is and U 🫵🏻 don’t know Y they call it DIC 🍆my culture (literally) is NOT ur costume
Scientists reported nearly doubling their productivity with AI in our study of more than 1,800 participants in the ChatGPT for Academic Researchers program.
Using privacy-preserving methods, we looked at how they use ChatGPT across their work—from developing ideas to analyzing data and writing code—and what it could mean for scientific discovery.
openai.com/index/science-at-…
10
442
Jacob Boysen retweeted
The code of life is written in nucleic acids, but proteins carry out the instructions. New work from @MSKCancerCenter led by the brilliant @LauraCKida introduces sctRNA-seq, a new single-cell assay that defines how cells translate between these languages
biorxiv.org/content/10.64898…



























