Scientists from Stanford and UCLA just found a hidden "off switch" for heart scarring. When the heart is stressed, the cells that cause dangerous scar tissue (fibrosis) get switched on by three chemical sensors working together. Blocking all three at once — but not just one — stopped the scarring in human heart tissue and in mice. This could become a new type of heart failure drug. science.org/doi/10.1126/scie…

Microbial community dynamics in a traditional Swiss mountain cheese over 142 years of cheesemaking "preserved cheese microbiome over 142 yrs" sciencedirect.com/science/ar…

Today, Sculpta releases the first demonstration of bioorthogonal barcoding ("bobcoding"), the process by which we: 1) covalently attach oligonucleotide barcodes to RNA at multiple internal positions with two-step chemistry, and 2) perform a multiplexed reverse transcription reaction to generate barcoded cDNA libraries with >99% barcoding accuracy. To the best of our knowledge, it is the first time ever that either of these steps have ever been performed by anyone. No one has ever broken the barrier of multiplexing prior to any enzymatic step in library preparation, driving simplicity and higher data quality. Instead of other methods that add 0 or 1 barcode, we add one barcode every ~300bp, adding redundancy and fidelity to RNA measurements. In our first application described in our pre-print below, we used bobcodes to create the first RNA isoform-resolved drug screening platform. It beats Novartis' DRUG-seq platform by: - generating full length transcript capture of RNA isoforms, instead of just 3' end counting - capturing 25-fold more RNA splicing events - reducing barcode swapping by 10-fold - using 18 fewer PCR cycles (~250,000 less amplification) - reducing sample-to-sample variability in gene expression measurements - eliminating costly and cumbersome library fragmentation/tagmentation steps completely - reducing workflow complexity and number of steps - reducing overall protocol duration by ~25% Though our chemical barcoding method improved transcriptomic data quality while also being simpler and faster than all existing methods, the implications of this work go well beyond this particular use case: The gate to greater applications of AI in transcriptomics is not a lack of compute. Nor is it a lack of data volume. The elephant in the room has always been our limited ability to faithfully and accurately measure cellular RNAs. Sculpta now has line of sight to build what we call the first ground truth transcriptomics platform [for single cell, spatial and more] for the future of biological research, drug development and AI-enabled discoveries for the betterment of human health and longevity. [PS Until the kind folks at bioRxiv get through the apparent backlog of AI slop submissions, Sculpta will host the pre-print PDF on our website. You can sign up for product offerings and other updates at this link too] sculpta.bio/preprint

Congrats to @viswacolluru @augustallen @brieski and the whole @enveda crew!

AI is getting cheaper more quickly than any other transformative tech in history. At a given level of performance, cost has fallen ~47%/quarter since 2023. That’s 4× faster than DNA sequencing, 6× faster than compute, 18× faster than lithium batteries, and (up to 1973) 54× faster than electricity.

This is cool — of the ~20,000 human proteins, 610 are the targets of approved drugs:

Wow! multiple sclerosis is caused by viral reactivation in monocytes and B cells months before clinical onset. nature.com/articles/s41591-0…

This is the real Turing test if you think about it

I don’t care about Navier-Stokes. I want to see AI solve the Voynich manuscript.

The protein measurements revealed cell-type-specific proteome architecture -- shaped in part by protein stability, and complex coordination -- that was undetectable in our mRNA data. Specifically, protein covariation within a cell type suggests cell-type-specific protein-protein interactions and functional rewiring of biological pathways independent of abundance differences. Some examples: 1️⃣ Arp2/3 complex shows distinct protein-level coordination across immune cell types, while this organization is largely absent from the corresponding mRNA data. 2️⃣ Protein covariation reveal an axis of translational states in B cells that tracks inversely with the abundance cytokine GDF6, suggesting reciprocal regulation of protein synthesis and GDF6 priming. The broader message is simple: 🔷 Single-cell proteomics opens a new window to a layer of functional coordination that had remained remained undetected. 👉 We are excited to open this window wider and investigate deeper ! This project was capably led by Luke Khoury and enabled by enjoyable collaborations with @bruker @BrukerMassSpec, Jeff Mold and team at @karolinskainst. Thanks to @MALifeSciences, @AllenInstitute, and @NIH for funding this research! 3/

for the future: in the wake of _____ results, it kind of looks like we've built a ____x or ______x Aristotle ...

Also the technology to even measure most of it at a reasonable scale doesn't exist yet.

This is notable. DeepSeek, a lab usually first to pioneer novel algorithms and architectures, is saying that at this point, the ROI of improving data quality far exceeds that of working on novel post-training algorithms. I think this has already been true for some time for non-lab practitioners. If you're doing llm post-training, 80% of your effort should go into looking at your data. This means: - Hiring experts to dig through your RL tasks - Sifting through rollouts and sft data by hand to remove suspicious samples. Make sure all tasks are actually passable. - Making sure your data is diverse in both difficulty and category.

N = 2

I literally had the debate at a party recently. Pretraining model size scaling is dead because transformers are saturated. Only gains now are data. The group found it to be a very controversial statement 😂

I was looking for an old tweet and found this. Jfc @geochurch is an alien from the future… and @jshendure lab are like the Men in Black.

Side note: when we released ARC-AGI-3 in March, and frontier models scored <1% on it, a few Singularitarian poasters took it as a personal insult, and got very worked up about it. They argued the benchmark was fundamentally broken, that it could not even be solved by the smartest humans, that the max reachable score was actually 40%, etc. We had to deal with a torrent of insults and hate poasts since because we had released an unsaturated benchmark. As it turns out, the benchmark is perfectly calibrated. It is straightforward for a human to score 100% if they do better than average people – all you need is to use fewer actions than our human baseline (which is not a strong baseline, as we used unfiltered human testers). And naturally as a result it's also very feasible for AI to score 100% once real progress towards agentic general intelligence has been made. The trajectory of AI from <1% to 100% over the course of 6 months shows that the benchmark was able to snapshot the recent rise of agentic capabilities. And that rise has happened faster than most people expected, including us.

Mechanics of compression-driven morphogenesis Read this Review from our special issue #DevSIextracellular by Jue Yu Kelly Tan and Chii Jou Chan. journals.biologists.com/dev/…

Anthropic shipped Mythos 5.1 today, leading the ProteinGym benchmark at 49.3% rank correlation against real lab measurements. Two weeks ago they published Claude running autonomous protein binder design campaigns at a 27% hit rate versus the 10-15% that's typical today. Both results are impressive, but both point to a problem most people aren't (yet) seeing. Start with the shape of that ProteinGym chart: Mythos 5.1 - 49.3% Opus 5 - 47.7% Mythos 5 - 45.8% Gemini 3.1 Pro - 37.0% Sonnet 5 - 36.6% GPT-5.6 - 35.5% Six frontier general-purpose models inside a 14-point band. None of them dedicated protein models. All trained on roughly the same public corpus. That's not anyone achieving a moat; it's a a floor rising. What ProteinGym actually measures: ~217 deep mutational scanning assays covering millions of variants, where every one of them is an experiment somebody already ran and published. Predicting results that exist is a different problem from generating results that don't. The binder campaign has the same shape, and Anthropic says so in its own technical report. Co-folding confidence turned out to be a useful filter but not a guarantee, and "experimental screening remains the only way to learn" which targets a campaign actually succeeded on. Adaptyv, who ran the wet lab, called it an open-loop experiment and said the next step is closing the loop. You can't close that loop at 30 designs per target with a multi-week CRO turnaround. You get one shot and no statistical power to learn anything from it. In contrast – last March, @ManifoldBio and @NVIDIA tested 1,000,000 designs against 127 targets in a single multiplexed experiment, measuring over 100 million protein-protein interactions. Roughly 750x the design count of the Anthropic campaign, in one run. (disclosure: I'm on Manifold's board, and an investor.) The model they were validating in that study was Proteina-Complexa, which is one of the ten open-source design models Claude used in the Anthropic campaign. Design is converging on a shared toolkit, and ... commoditizing. Measurement is not. Which brings me to the ladder below. Every public benchmark, and both of Anthropic's recent headline results, live near the bottom of it. Expression. Binding. In vitro affinity. There is no ProteinGym for biodistribution and for what actually translates into human biology (and into therapeutics that work). Not because nobody wants one, but because the data doesn't exist at benchmark scale. Someone has to generate it, at billion-scale. Whoever does has the data that matters to train the models that will actually make a difference. My bet is that the first to million-scale will be the first to billion-scale.

It’s September 1st so here’s your reminder of what Utah will look like in a couple weeks 😍