We sample from our generative biological sequence models at peta-scale. Read about our work at https://xitter.cf/t.co/j5R4jau8qI
BOS | BSL | OSL | KBH | LGA
Joined August 2020
- Tweets207
- Following13
- Followers1.2K
- Likes493
RT @lizbwood: This paper (@KasperKrunderup and @EliWeinstein6) is a tour de force is experimental codesign, and shows exponential speedups…
This quoted post is unavailable.
JURA Bio, Inc. retweeted
It's been an exciting week at @jura_bio!
We got to share a significant update to Variational Synthesis. cVS gives us more control and flexibility in designing and synthesizing large-scale DNA libraries.
As always, @lizbwood wrote a great intro!
Introducing: continuous variational synthesis
📄 jurabio.com/s/cvs.pdf
We’re pleased to announce a new advance in our ability to synthesize generative model-designed DNA sequences at petascale.
JURA Bio, Inc. retweeted
First you could train an in vitro generative model – now, thanks to our guest star player @AlanNawzadAmin, you can fine-tune it as well!
Introducing: continuous variational synthesis
📄 jurabio.com/s/cvs.pdf
We’re pleased to announce a new advance in our ability to synthesize generative model-designed DNA sequences at petascale.
JURA Bio, Inc. retweeted
A common misunderstanding. Instead, the model can capture full conditionals -- and then sample in DNA. If the Ab CDR3 example is too confusing (because of the apparent flexibility of that region), then look to the enzyme & human peptide examples from both papers.
JURA Bio, Inc. retweeted
To be fair to you -- this is a criticism we get a lot. But I have never understood it. To me it's like saying:
You developed a 2,000,000,000-megawatt reactor at 1-trillionth the old cost of energy! that's 100x the total world's energy needs! Why'd you do that?
JURA Bio, Inc. retweeted
That's right. It's just like -- let's design 10^16 sequences and have them all be made.
What has always been surprising to me is when these realized designs outperform in-silico-only models, as in the antibody CDR3 (vs. IgLM) task in rdcu.be/e8zUb.
JURA Bio, Inc. retweeted
Very pleased to share that arxiv.org/abs/2512.15984v1 -- a method to pack and extract more information from high throughput experiments, using variational synthesis -- was accepted at NeurIPS.
You can also read the blogpost: jurabio.com/blog/lift
JURA Bio, Inc. retweeted
Also @AlanNawzadAmin is one of the most creative computational scientists in bio. Love his work.
JURA Bio, Inc. retweeted
This looks phenomenal! I think Jura Bio's tech is under appreciated.
Introducing: continuous variational synthesis
📄 jurabio.com/s/cvs.pdf
We’re pleased to announce a new advance in our ability to synthesize generative model-designed DNA sequences at petascale.
JURA Bio, Inc. retweeted
A few years ago, @EliWeinstein6 and @jura_bio wrote the output of a DNA synthesizer as a probability distribution – a family of generative models that you can sample from in real life, Variational Synthesis (VS). But fitting VS models is brutal. Now we’ve fixed that! 1/6
We were fortunate to have @AlanNawzadAmin spearhead this effort. Please check out his tweetorial below:
A few years ago, @EliWeinstein6 and @jura_bio wrote the output of a DNA synthesizer as a probability distribution – a family of generative models that you can sample from in real life, Variational Synthesis (VS). But fitting VS models is brutal. Now we’ve fixed that! 1/6
Today we're introducing continuous variational synthesis.
Spearheaded by @AlanNawzadAmin, cVS brings to manufacturing-aware architectures more of what we expect from modern generative model architectures, including pretraining and fine-tuning, and flexibility wrt hardware.
Introducing: continuous variational synthesis
📄 jurabio.com/s/cvs.pdf
We’re pleased to announce a new advance in our ability to synthesize generative model-designed DNA sequences at petascale.
JURA Bio, Inc. retweeted
I am once again urging everyone to use variational synthesis for problems like this. (Thanks for citing us, Bruce!)
JURA Bio, Inc. retweeted
I am grateful that Chinese coverage of an American company is worth reporting and sharing, @SynBioBeta. But why not cover us yourselves sometime?
JURA Bio, Inc. retweeted
At long last, SynBioBeta decided to report on @jura_bio… or at least share the reporting that has happened in China. For those who’d like to read about our work but are stuck with English:
jurabio.com/blog/scalinglaw
@jura_bio reports a biological-AI scaling law built through lab-generated data. Its system screened a large sample from ~10¹⁶ antibody designs against 100 pHLA targets, with model performance improving predictably as data grew. Findings currently cover a specific antibody–pHLA task.
mp.weixin.qq.com/s?__biz=MzI…
Subscribe to our newsletter and get the biggest biotech news straight to your inbox 🧬
syntheticbiologysummit.com/?…
Read our technical blogpost: jurabio.com/blog/scalinglaw
@jura_bio reports a biological-AI scaling law built through lab-generated data. Its system screened a large sample from ~10¹⁶ antibody designs against 100 pHLA targets, with model performance improving predictably as data grew. Findings currently cover a specific antibody–pHLA task.
mp.weixin.qq.com/s?__biz=MzI…
Subscribe to our newsletter and get the biggest biotech news straight to your inbox 🧬
syntheticbiologysummit.com/?…
Thanks for highlighting our work, Peter!
A lot happened in AI x bio over the past week. Here’s what you might’ve missed. 🧵 (1/8)
☑️ Demis Hassabis stepped back from running NanoBanana to go all in on science, AGI and Isomorphic. And one explosive report says he actually wanted to leave Google.
☑️ Meanwhile, four Google legends actually did leave - to build a startup that aims to automate scientific discovery.
☑️ The AI system Biomni spent five days doing months of biology-model research.
☑️ AI found a hidden chemotherapy benefit inside a failed cancer trial.
☑️ Jura Bio says it found a scaling law for biological AI.
☑️ The interpretability masters at Goodfire used a protein model to predict not only whether 2.1 million mutations may be harmful, but what they may break and why.
JURA Bio, Inc. retweeted
Nice! I think adding the linear baseline actually makes things MORE compelling because there is a flattening out of performance that your transformer based method seems impervious to.
So in some ways you discovered another scaling law but for linear models.
I wonder if the hype around linear baselines was simply that those evals were overdetermined and still in the regime of comparable performance!
JURA Bio, Inc. retweeted
On model ablations: Linear models plateau on this data (dashed), even with foundation model representations (colors). We needed big transformers (solid) to scale.
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.
JURA Bio, Inc. retweeted
There's been a lot of talk about data scaling in bio lately but in a kind of underbaked way, as if just measuring more "stuff" will solve everything. Interesting to see a nice example in practice!
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.