David Li

Had fun solving the cryo structure of Evo-Φ36 with the awesome team of @samuelhking, @driscoll_cl, and @maxewilkinson! Samuel is a beast and the reason there is a 'skevophage' react on Stanford slack.

❓How many qubits are needed to break cryptographic systems🔐, e.g., ECC-256 in BTC? With a new fault-tolerant quantum architecture arxiv.org/abs/2603.28627, ECC-256 can now be broken with ~10k atomic qubits. The largest atom array has 6k+ atomic qubits: arxiv.org/abs/2403.12021

Today, we introduce Oratomic. We are on a focused mission to build the world’s first fault-tolerant quantum computers and unlock their transformative applications. Quantum computers offer a fundamentally new way of understanding and interacting with the physical world. Our recent scientific advance finds that Shor’s algorithm is possible with as few as 10,000 reconfigurable atomic qubits: arxiv.org/abs/2603.28627 Our team integrates world-class expertise in quantum error correction, neutral atom systems, artificial intelligence, and optical engineering. We are working together to make fault-tolerant quantum computing a reality. To learn more about Oratomic and our team, visit oratomic.com

Big news: the thymus may be critical for adult health What if we missed something fundamental about the immune system? In two back-to-back papers in Nature, we show thymic health (AI on 30,000+ CT scans) links to longevity, disease risk & immunotherapy outcomes.

I am so excited to share our new paper in @Nature: the first programmable, site-specific integration of a large DNA payload into T cells in vivo. A single IV injection results in therapeutic levels of TRAC-targeted CAR T cells in multiple models. #Ack1" target="_blank" rel="nofollow noopener">nature.com/articles/s4158… a 🧵

The next frontier in AI is science, the physical world. Biology will be the *most* impactful and consequential application of AI, from curing disease to biosecurity. We need AI labs pushing both the limits of biological design and the systems to defend against it. We’re building exactly this intelligence layer and scaling fast. Hiring across AI and biology. Come build at the frontier, let's talk. job-boards.greenhouse.io/radicalnumerics

Great to the see the flurry of single gene knockdown Perturb-seq like atlases from cell-lines, mouse brain etc over the last few days. These are undoubtedly very valuable datasets. I just want to re-iterate a few other very important expt. design considerations 1/

📢 We’re launching Proteina-Complexa — and after the Jensen keynote mention, we definitely had to post this thread now ;) Atomistic binder design with generative pretraining + test-time compute, plus large-scale wet-lab validation. Project page: research.nvidia.com/labs/genair/pr… 🧵 1/n

I'm excited to announce some major updates to our ProteinEBM paper with Chenxi Ou and @sokrypton!

Check out this AMAZING work from Claire and Ravi. Fish have midlife crises too (okay 'discrete, sharp, and stable behavioral transitions')

Aging may feel gradual… but what if it’s not? In our paper out today, we tracked fish continuously from puberty until death. This gave us a unique view of how aging unfolds across the adult lifespan. 🧵

From the neuroscience perspective, behavior is unmatched as noninvasive quantitative readout of brain function (and best of all, it is a readout that matters). This initial lifelong continuous screen of behaving vertebrates provided a wealth of new ideas regarding progression and staging of brain function across the adult lifespan. Truly impressive work from Claire, Ravi, and the whole creative and dedicated team. Lifelong work takes a long time, even with killifish!

Axiom launched six months ago with one conviction: mathematics is the right foundation for building systems that reason. Today we announce Axiom's Series A. We raised $200M at a $1.6B+ valuation, led by @MenloVentures, to extend our lead in formal mathematics into Verified AI.

Our live tissue clearing paper is out in @naturemethods! We achieved optical clearing of mammalian brain tissues without compromising normal neuronal function. Big congrats to @Shigenori774 and our wonderful collaborators! 🎉 nature.com/articles/s4159… (1/10)

Thrilled to announce alphagenome-pytorch, an accurate, readable, and careful port of AlphaGenome's architecture and weights to PyTorch. Work with @gtcaa @m_kjellberg @chriswzou @tuxinming as part of the GenomicsxAI initiative between @anshulkundaje and @pkoo562 labs.

🚨Preprint! From my PhD “Genome delivery of a contractile tailed phage and its superinfection exclusion mechanism”. We use #cryoEM to study the genome ejection of the #phage T4, revealing how the tape measure protein regulates the process @NMITaylorLab biorxiv.org/content/10.648…

To make Evo 2 more accessible, we're releasing Evo 2 20B, a checkpoint that achieves 40B-level performance on a single H100, as a drop-in replacement. This came out of model surgery with @danielchang2002, and we are excited to see people build on it! github.com/ArcInstitute/e…

Evo 2 is out today! Check out the new 20b and Garyk's thread! Excited by the potential of genome language models for enhancing and expanding through design our understanding of biological sequence space

🚨 Today in @Nature, we report GEMINI—a genetically encoded intracellular memory device that writes cellular dynamics into tree-ring-like fluorescent patterns within cytoplasmic protein assemblies.[1/n] nature.com/articles/s4158…

Protein–protein interactions (PPIs) are key to discovering and interpreting new biological functions. We’re excited to introduce 𝑭𝒍𝒂𝒔𝒉𝑷𝑷𝑰: a new application of gLM2 that uses genomic language modeling to predict proteome-wide PPIs in microbial genomes in minutes.

OK. Here it is. Over the last year, the @OctantBio and OpenADMET team have been hard at work developing scalable, quantitative, data-rich, and low-cost methods for assessing CYP reactivity and inhibition. The interplay between building a data engine and building predictive models is often the most subtle, difficult, and impactful work in small-molecule AI/ML. The blog post below is intended to highlight these issues and serve as a dialogue starter to help us, help you. In the post, we are: 1. Give some background on the types of assays we are building and the technologies we are developing to scale ADMET datasets. We also go deeper into the tradeoffs inherent in building assays and exposing some of our design decisions. 2. A data drop of some of the largest self-consistent datasets for CYP reactivity and inhibition (CYP3A4 & CYP2J2). Importantly, we are exposing the raw datasets and are urging the community to help us design better methodologies and analytical tools to best extract the most informative data. This is a teaser dataset for the competitions we are running on CYP reactivity/inhibition blind challenge later this year. 4. A call to the AI/ML and ADMET community to help us decide on the types of data we should be collecting and holding blind competitions for. Should we focus more on inhibition or reactivity? What about TDI, metID, microsomal, and other types of assays? What should our screening funnel look like? What summary statistics should we try to predict? How useful is the raw data and uncertainty? How important is true negative data, or is it more important to get more quantitative data? What compounds should we screen? Given a budget, what data should we collect (some assays are more expensive than others)? How should we split the data for the blind challenge?