Andrei Slabodkin

ML Scientist at @jura_bio

[Sasha Malysheva @aimalysheva]

meet @mostik_ai!

what happens when you put 12 PhDs in one room for four months? first place on the ARC-AGI leaderboard, which I can't say much about while the competition is still running. and this, which I can.

everyone's arguing about whether open models will catch up to frontier models. we think it's the wrong question. here's the one we pose: why does a frontier model have to generate your answer at all, when the only thing you need from it is the reasoning?

we do this by enabling models to communicate in latent space. through our protocol, hidden states pass straight from a frontier model into a small one running on your infrastructure -- no text between them, and neither model is fine-tuned. two models from different families, sharing reasoning, both left untouched.

how do we know it works? we tested it on a setup where a 753B model reads the problem, and a 4B edge-class model writes the answer. with this approach, we get results 80% as accurate as the frontier model, but at 20x faster performance.

we're committed to preventing frontier model lock-in and are already partnering with inference providers to accelerate open-weight adoption. we've done this between 15 of us, in four months, 12 PhDs and a Fields medalist, backed by @generalcatalyst

WIRED has the first external account of the company and the work: https://t.co/tP8nItCsDl
full writeup, the setup, and all the numbers: https://t.co/C9NZ5vtV1V

[Andrei Slabodkin @slabodkin]

A new scaling law just dropped – for your (and mine) favorite modality!

[Elizabeth Wood 🧬🖥️🥼 @lizbwood]

Can we achieve the same scaling laws in biological AI as we have in the rest of machine learning?

We found that @jura_bio's mix of designed data generation and training produce robust scaling laws: more data reliably lead to better predictions over many orders of magnitude.

[Andrei Slabodkin @slabodkin]

@suchenzang In other words, come to biology! 🙃

[Andrei Slabodkin @slabodkin]

When you work in ML, there are these really rare moments when - while you are still sceptical - the magic happens, and the model just works.
Antibody binding prediction that generalizes to unseen epitopes, see below

[Elizabeth Wood 🧬🖥️🥼 @lizbwood]

https://t.co/tLOOOGQKIU

[Andrei Slabodkin @slabodkin]

@KevinKaichuang Or, "can a radio fix a biologist?"

[Andrei Slabodkin @slabodkin]

If you ever wondered what a "binding landscape" looks like (from the epitope point of view), now is the time to go and check it out

[Elizabeth Wood 🧬🖥️🥼 @lizbwood]

We're launching our new target intelligence dashboard, MESA, with a nice interactive case study from our recent TCRm campaigns. 82M AI-designed scFvs manufactured and tested against 100 targets, simultaneously.

https://t.co/4sUJZ0CEUp

[Elizabeth Wood 🧬🖥️🥼 @lizbwood]

The race to first-in-human is accelerating globally. The obvious response is to scale up: more automation, more cells, more sequencing, more capital. This will work until it doesn't.

At @JURA_Bio, we've been building beyond the obvious.

[Andrei Slabodkin @slabodkin]

@owl_poster Does the number 239 ring any bells?

[Andrei Slabodkin @slabodkin]

Who would like to synthesize the whole Observed Antibody Space at a cost of several MacBooks?

[Eli Weinstein @EliWeinstein6]

I’m thrilled to announce: at @jura_bio we’ve constructed a new kind of generative protein model, which allows us to manufacture its generated designs in the real world at petascale.
Paper: https://t.co/qgxfNPP30K
Blog: https://t.co/An1USoCNoz

[Corey Watson @ctwatson29]

https://t.co/D8ErVOpVzq

[Alex Rives @alexrives]

We have trained ESM3 and we're excited to introduce EvolutionaryScale.

ESM3 is a generative language model for programming biology. In experiments, we found ESM3 can simulate 500M years of evolution to generate new fluorescent proteins.

Read more: https://t.co/iAC3lkj0iV

[Andrei Slabodkin @slabodkin]

@lizbwood @notSoJunkDNA Madeira! It's like Iceland but with (sub)tropical forests

[Andrei Slabodkin @slabodkin]

@PRobertImmodels @DJGould94 I still think the same when with respect to myself though (please, remind me of this if I start calling myself a doctor all the time when I finally defend)

[Andrei Slabodkin @slabodkin]

@PRobertImmodels @DJGould94 I used to think the same (I don't trust people who sign as "Someone Someonesson, PhD" everywhere, why would you do this?) – but then I heard, on multiple occasions, that for people from underrepresented backgrounds the title may be much more important

[Khang Le Quy @KhangLeQuy]

(1/7) Curious about what a blood sample can reveal about humoral immunity? Our latest work combined sequencing methods and mass spectrometry to explore BCR and antibody repertoires and assess their synergistic potential. https://t.co/kGQEgTm93q

[Pieter Meysman @chevaliersf]

The IMMREP23 @kaggle #TCR-#epitope specificity #MachineLearning #prediction challenge is live now! Spread the word! @airr_community @iRecRepPapers
https://t.co/lDMxM2jOW0

[Maria Chernigovskaya @mchernigovskaia]

(1/8)🎉A fresh preprint in which we present LIgO — a powerful tool to simulate adaptive immune receptor (AIR) and repertoire (AIRR) data for the development and benchmarking of AIRR-based ML 🧵⬇️
https://t.co/0iif0WbUP2

[David Gfeller @GfellerD]

https://t.co/duAWlQXQeC. A good reminder that predictions of TCRs recognizing 'unseen' epitopes (i.e., epitopes without any known TCR) remains an unsolved challenge.

[Andrei Slabodkin @slabodkin]

@Will_ig_ab @SandveGeir Thank you, William! Of course, questions and feedback are highly appreciated :)

[Andrei Slabodkin @slabodkin]

1/8 New preprint: generative modeling of AIRR repertoires, the last piece of my PhD, >2 years of work, a project that is very dear to me https://t.co/gwZt4WBuLY

[Andrei Slabodkin @slabodkin]

8/8 We thank @HelmsleyTrust, @forskningsradet, @UniOslo, @UiO_LifeSci, @EU_H2020, @ireceptor_plus, Norwegian Cancer Society, @JCoDiRC, Innovative Medicines Initiative 2 Joint Undertaking for generous funding.

[Andrei Slabodkin @slabodkin]

7/8 Huge thanks to Ludvig M. Sollid, @SandveGeir, and the co-last authors @PRobertImmodels and @victorgreiff