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.

On model ablations: Linear models plateau on this data (dashed), even with foundation model representations (colors). We needed big transformers (solid) to scale.

Elizabeth Wood, PhD@lizbwood.bsky.social · 2mo ago

Can we achieve the same scaling laws in biological AI as we have in the rest of machine learning? We found that @jura.bsky.social's mix of designed data generation and training produce robust scaling laws: more data reliably lead to better predictions over many orders of magnitude.

Can we achieve the same scaling laws in biological AI as we have in the rest of machine learning? We found that @jura.bsky.social's mix of designed data generation and training produce robust scaling laws: more data reliably lead to better predictions over many orders of magnitude.

Overall, I am excited about geometric causal models' potential for bringing causal machine learning methods and ideas into new areas of science, including especially the molecular sciences where symmetries abound.

Preprint: arxiv.org/abs/2607.051.... This is joint work with David Blei @bleilab.bsky.social

Geometric Causal Models

Scientists often seek to draw causal inferences from structured data that is not independently and identically distributed, such as spatial data, network data, or molecular data. We develop geometric ...

arxiv.org

Our new preprint pushes causal machine learning into new scientific domains using probabilistic symmetry and geometry.

We are hiring at @jura.bsky.social both on the ML team (worldwide) and the wetlab team (Boston). Please get in touch if you've been thinking about it. ⏩ careers@jurabio.com

Manufacturing-aware generative models enable petascale synthesis of designed DNA - @lizbwood.bsky.social @jura.bsky.social go.nature.com/3NxXt1I

Manufacturing-aware generative models enable petascale synthesis of designed DNA - Nature Biotechnology

A generative modeling framework enables large-scale designed DNA synthesis, yielding candidate cell therapies.

go.nature.com

Ever wondered what it'd be like to run a de novo antibody campaign against 100 of the hardest targets, simultaneously, with 76% success rate, and have the complete wetlab validated results days from the project's start? We show you what that looks like, too: www.jurabio.com/mesa

Elizabeth Wood, PhD@lizbwood.bsky.social · 9mo ago

JURA is launching a new target intelligence dashboard called MESA jurabio.com/mesa with a nice interactive case study from our recent TCRm campaigns. Go check it out -- we'll open to crowd-sourced targets soon. @jura.bsky.social

JURA is launching a new target intelligence dashboard called MESA jurabio.com/mesa with a nice interactive case study from our recent TCRm campaigns. Go check it out -- we'll open to crowd-sourced targets soon. @jura.bsky.social

For most of the history of molecular biology, the experimental bottleneck has been so severe that we've learned to work around it rather than through it. The workaround is prediction. If you can only physically test a small number of molecules, you need to be good at guessing which ones to make.

We are at the #StartupVillage at #EurIPS today — come say hi 👋 @jura.bsky.social

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 until it doesn't. At @jura.bsky.social we've been building approaches beyond the obvious.

Our LeaVS preprint is up! We demonstrate empirically & prove theoretically that LeaVS can dramatically accelerate learning, increasing the effective dataset size by orders of magnitude. Really useful biological assays have intractable bottlenecks: make the most of what you can get. @jura.bsky.social

Eli Weinstein@eliweinstein.bsky.social · 11mo ago

We're excited to present LeaVS, a method to scale up learning for protein function models. It is based on the co-design of wet lab experiments and in silico training.

For every Nobel that goes to a criminally under-recognized woman scientist (Brunkow, Karikó), or fails to go (Candy Lee), a week of mourning and reform for an academic system wherein you can do Nobel-prize-worthy-work and still end up without a conceivable path to being a professor.

Nature@nature.com · 12mo ago

BREAKING: The Nobel Prize in Physiology or Medicine has been awarded jointly to Mary E. Brunkow, Fred Ramsdell, and Shimon Sakaguchi "for their discoveries concerning peripheral immune tolerance" Stay tuned for more. #NobelPrize

Let's say all of a sudden you found yourself with 100,000+ wetlab validated, de novo AI designed, fully human GLP1R binding antibodies? Could you turn the patent suite into a financial instrument? Anyone who thinks about this you could connect me to?

How do you train AI on protein data when functional sequences are astonishingly rare? Traditional iterative optimization fails because most measurements show zero. Our newly announcing learning system, LeaVS, shows a way.

Elizabeth Wood, PhD@lizbwood.bsky.social · last yr.

You can read more in our post at www.jurabio.com/blog/leavs; preprint forthcoming. @jura.bsky.social @eliweinstein.bsky.social @mgollub.bsky.social @highvariance.bsky.social

A fundamental lesson of modern AI is that scale is essential: training bigger models on bigger datasets unlocks new capabilities. A fundamental lesson of AI engineering is that scaling up isn't trivial: it is not just a matter of spending more money and resources.

🚨 One week left to apply! I am hiring a fully funded PhD fellow and a postdoc to work on membrane protein structural biology and pharmacology at Univ Copenhagen Come work with a great team and exciting projects in a collaborative environment! 🔗 Links to the ads in comments #CryoEM #AcademicJobs

I'm looking for my first PhD student! We will push the frontiers of probabilistic machine learning for the molecular sciences, and study how to design new algorithms that exploit the unique properties of molecular systems to learn about the world. efzu.fa.em2.oraclecloud.com/hcmUI/Candid...

PhD scholarship in Machine Learning for Molecules - DTU Chemistry

Advance large scale data generation for chemical and biological AI in a 3 year PhD. Work on the frontier of active learning, and develop novel probabilistic machine learning techniques for experiment ...

efzu.fa.em2.oraclecloud.com

Want to work at the frontier of active learning and AI-guided experimentation? @eliweinstein.bsky.social is hiring PhD students at @dtu.dk DTU Chemistry. Full job link: efzu.fa.em2.oraclecloud.com/hcmUI/Candid... Thanks for re-sharing widely!

PhD scholarship in Machine Learning for Molecules - DTU Chemistry

Advance large scale data generation for chemical and biological AI in a 3 year PhD. Work on the frontier of active learning, and develop novel probabilistic machine learning techniques for experiment ...

efzu.fa.em2.oraclecloud.com

Totally agree that standard biological data isn't built for AI. We need to think strategically about how we build better, less biased richer datasets specifically for machine learning

Elizabeth Wood, PhD@lizbwood.bsky.social · last yr.

The biggest challenge for AI in biology isn't just models, it's the data used to train them. Standard biological data isn't built for AI. To unlock generative AI for drug discovery, we must rethink how we generate and capture data. 1/

Thrilled to demonstrate VISTA, our platform for creating high quality, AI-ready biological data loops at massive scale!

Elizabeth Wood, PhD@lizbwood.bsky.social · last yr.

The biggest challenge for AI in biology isn't just models, it's the data used to train them. Standard biological data isn't built for AI. To unlock generative AI for drug discovery, we must rethink how we generate and capture data. 1/

Great coverage of VISTA by the wonderful @faylinphd.bsky.social over at GEN: www.genengnews.com/topics/artif... Read about VISTA: www.jurabio.com/blog/vista

Jura Bio’s AI Data Loop Enables Large-Scale De Novo Antibody Design

VISTA, an AI-controlled data loop system, guides large-scale antibody design from scratch for applications in cancer therapy.

genengnews.com

When AI drives your data generation, learning is more efficient and effective. Take a deep dive into VISTA:

Elizabeth Wood, PhD@lizbwood.bsky.social · last yr.

The biggest challenge for AI in biology isn't just models, it's the data used to train them. Standard biological data isn't built for AI. To unlock generative AI for drug discovery, we must rethink how we generate and capture data. 1/

So excited about this - we did iterative generative design at large scale with variational synthesis, and got human scFv candidates against some of the hardest therapeutic targets around.

Elizabeth Wood, PhD@lizbwood.bsky.social · last yr.

The biggest challenge for AI in biology isn't just models, it's the data used to train them. Standard biological data isn't built for AI. To unlock generative AI for drug discovery, we must rethink how we generate and capture data. 1/

The biggest challenge for AI in biology isn't just models, it's the data used to train them. Standard biological data isn't built for AI. To unlock generative AI for drug discovery, we must rethink how we generate and capture data. 1/

We can make population genetics studies more powerful by building priors of variant effect size from features like binding. But we’ve been stuck on linear models! We introduce DeepWAS to learn deep priors on millions of variants! #ICML2025 Andres Potapczynski, @andrewgwils.bsky.social 1/7

This week is the BIO conference in Boston's Seaport. If you are in town for it and are interested in meeting to learn more about @jura.bsky.social please feel free to send a DM! Would love to see you.

“The search is on for 25 researchers from around the world in Lund University’s largest ever international recruitment drive. The main focus will be on AI research and other areas in which the University has a strategic focus.” www.lunduniversity.lu.se/article/lund...