←
John Yang177 posts
John Yang avatar

John Yang

@johnyang100

Co-founder @ Reticular (YC F24). CS & Math @ MIT

◷ Joined November 2020▧ 13 Photos and videos
John Yang avatar
John Yang@johnyang100·13 Mar 2025

If you’ve ever - thought AI protein folding is magical ✨ - wanted more than a pLDDT score 🔎 - or just think mech interp in bio is cool 🤓 then read the 🧵 👇 on our first paper towards interpretable protein structure prediction just accepted to workshops at ICLR

A First Step Towards Interpretable Protein Structure Prediction With SAEFold, we enable mechanistic interpretability on ESMFold, a protein structure prediction model, for the first time. Watch @NithinParsan demo a case study here w/ links for paper & open-source code 👇

0:42
John Yang avatar
John Yang@johnyang100·10h

mit ai for bio 🚀🚀🚀

We’ve raised $110M to bring AI-powered biomolecular design to every scientist! Boltz’s Series A is led by @BessemerVP, following 50-fold growth in molecular token usage from our users since June and partnerships with 6 of the world’s 20 largest pharma companies.

0:16
John Yang avatar
John Yang@johnyang100·Nov 11

alphagenome 2 is great work. but surprisingly it only ever sees one genome predicting who gets heart disease or prostate cancer requires training on millions of genomes hoping alphagenome 3 trains on more than one

AlphaGenome 2 is live! A step change in performance and new capabilities over AG1! @adamgayoso's thread covers the key results. Manuscript: storage.googleapis.com/deepm… Blog: deepmind.google/blog/alphage… API: github.com/google-deepmind/a… A few notes under the hood & what's next:

John Yang avatar
John Yang@johnyang100·Nov 11

100% agree for humans born 5-10 yrs from now

NIK@ns123abcNov 10

Anthropic CEO predicts cancer, Alzheimer's, heart disease, mental illness will be cured in less than 10 years ACCELERATE

1:04
John Yang avatar
John Yang@johnyang100·Nov 10

scaling laws for human genome pretraining soon

Post attachment

Genomes are long but redundant. Not great for pretraining. To fix that, we've developed a lossless tokenization strategy that shortens sentence length by >2000x. For pretraining, we achieve a 0.55 validation CE on All of Us. Public replication on 1KG linked below.

0:48
John Yang avatar
John Yang@johnyang100·Nov 10

lossless compression for human genomes? richard hendricks would be proud

Genomes are long but redundant. Not great for pretraining. To fix that, we've developed a lossless tokenization strategy that shortens sentence length by >2000x. For pretraining, we achieve a 0.55 validation CE on All of Us. Public replication on 1KG linked below.

0:48
John Yang avatar
John Yang@johnyang100·Nov 10

the failure of Lp(a) drugs shows why we need interpretability in drug discovery. heart disease is heritable, but thousands of genes interact to cause heart attack. genomic language models learn these interactions. interpreting them will reveal new drug targets.

Massive failure at Novartis to improve cardiovascular outcomes by lowering Lp(a). What went wrong? What now? Is Lp(a) useful? lifespan.com/blogs/magazine/…

John Yang avatar
John Yang@johnyang100·Nov 10

fda finally allowing personalized gene therapies inspiring to see how far we've come since milasen

Congratulations to @kiranmusunuru & @AhrensNicklas for not only pioneering individualized gene editing for Baby KJ, but also helping shape the regulatory pathway for the patients who will follow, with implications far beyond one patient or one disease. The FDA's new draft guidance (in reply) is an important advance, recognizing that multiple patient-specific variants of a gene-editing therapy can be evaluated within a common regulatory framework, while maintaining rigorous standards for safety and efficacy. This is how science and regulation should work together.inquirer.com/health/fda-guid…

Post attachment
John Yang avatar
John Yang@johnyang100·Nov 10

bio sequence modeling solved soon

Pretty striking how much better 5.2 is at Pfam classification. It's clear that models are already natively reasoning over biological inputs and may soon exceed the capabilities of specialized models. We're going to need a lot of scalable, broad expt capacity in the next few years.

Post attachment
John Yang avatar
John Yang@johnyang100·Nov 6

psa: the "IQ" test these models use is dogshit. it's only 2 minutes long. the questions are 5th grade level. and the test-retest reproducibility is low. take it yourself & see. oqsweb.ndph.ox.ac.uk/webq/mo… all this marketing hype pisses me off. stains the whole field.

this Peter Thiel backed startup says its AI can find a 14 point IQ gap between your embryos it's called Nucleus and their tagline is literally "have your best baby" here's how it happens: > you do a round of IVF and end up with a bunch of embryos > you send Nucleus the DNA data for each one, up to 20 embryos > its AI reads more than 7 million spots in each embryo's DNA and compares them to what it learned from more than a million people > then each embryo gets scored on things like IQ, disease risk, hair color and eye color > you pick the one you want implanted scoring your embryos starts at about $6,000 every one of these embryos is naturally yours btw, with zero gene editing involved America's main medical genetics group says this is moving too fast with too little evidence but that hasn't slowed Nucleus down at all when the founder announced this new AI, he wrote "AI is rapidly getting smarter... Now, humanity can too." they even put ads on the NYC subway that said "IQ is 50% genetic" lol personally i feel kinda uneasy about this but IVF got called playing god too, and now nobody blinks at it so for better or for worse, i think this will soon be accepted as normal

Post attachment
David Lieb avatar
⟳ John Yang retweeted
David Lieb@dflieb·Nov 4

This could help families afraid to have kids due to genetic history finally feel able to. AI+genetics is going to be huge.

Couples with a family history of severe autism have been reaching out, hoping we could find a genetic cause to screen for. But the genetic signal was always sparse. Now our new work on rare variants & GPN-star has increased performance by over 50%. Another step towards ending genetic disease.

Post attachment
John Yang avatar
John Yang@johnyang100·Nov 4

the golden age of genomics is here. the rapid growth of whole genome sequencing data in the last 2 years has unlocked so many new methods. solving genetic disease is extremely soluble in intelligence & compute.

Couples with a family history of severe autism have been reaching out, hoping we could find a genetic cause to screen for. But the genetic signal was always sparse. Now our new work on rare variants & GPN-star has increased performance by over 50%. Another step towards ending genetic disease.

Post attachment
John Yang avatar
John Yang@johnyang100·Nov 4

lots of couples with severe autism in their family have been asking for this. excited to deliver! lots more ahead. we're really entering the golden age of genomics

This tweet is unavailable