1/ The need for a Global Alliance for Spatial Technologies arises from multiple driving factors, pushing the progress and use of spatial tech across various disciplines. #GESTALT

✨ Honored to step into the role of Vice Co-Chair of the Human Cell Atlas ✨ Working together with our new co-chairs Chloé Villani and Muzz Haniffa, alongside my fellow Vice Co-Chair Shyam Prabhakar. I can’t imagine a better next leadership team for the @humancellatlas. Beyond their strong track record, it’s a pleasure to work with such thoughtful and collaborative colleagues. This is a particularly meaningful moment for me, having been part of the HCA journey for almost a decade. I have had the privilege of co-chairing the HCA Technology Working Group and Industry Partnership Program, and more recently serving on the HCA 2.0 Task Force, helping shape the vision and strategy for this next phase. As we move towards HCA 2.0, we are defining the future vision and roadmap for the next generation of the Human Cell Atlas #HCANext10. In my new role, I will specifically focus on leading the HCA technology roadmap, working closely with our technology community as well as biotech and pharma partners to integrate the latest single-cell and spatial technologies into the HCA roadmap. I’m very grateful to the HCA community for the trust. Excited for the next chapter of the Human Cell Atlas and for what we can build together. @sangerinstitute @broadinstitute @ASTARsg @cnag_eu

Benchmarking with Honesty | S1 E1 The Life & Science Podcast 🫶 open.spotify.com/episode/2eO…

Still buzzing after #AGTA2026. ⚡ I spent the first part of the week at the #UNSW Roundhouse with the Australasian genomics community, and it was a good reminder of what makes this field worth being in. There was a PhD student I could see shaking a little before her first poster, and five minutes later, the room was firing questions at her like she'd been doing this for years. If you were there, you know the feeling in that room. That's the stuff you don't get from a paper. That was the week, really. A round corridor full of conversations that turned into collaborations. A whole room of people who light up over sequencing data. New science friends made over average coffee. The easy back-and-forth that drifts between life and science, and a bit of magic (yes, magic!), and back again. And, of course, the honest talk about where funding is really heading. And it's a good time to be in it. The tools are getting faster, the platform providers are willing to collaborate, the questions are getting bigger, and this community is the reason Australasia gets to help answer them rather than watch from the sidelines. Huge thanks to the organizers, who pulled off something that looked effortless and absolutely wasn't. Every session, the content, the food, the gala dinner in that magnificent venue, every well-timed coffee break. That's a lot of work behind the scenes. Thank you. And yes, the talks are brilliant, but they're not really why you come. You come for the people. You come to belong to something. The conference is only a few days a year. AGTA is the rest of it. I joined a few years ago to stay connected all year round: discounted registration (which alone can cover the membership fee), member-only content, and real support for students and early-career researchers through scholarships and prizes. Add the introductions, the collaborations, and being first to know what's coming, and the maths isn't close. You get back far more than you put in. The reason I joined is simple: this is my community, and I want to keep showing up for it. 👉 Join or renew: lnkd.in/gugEPdeb Australasian Genomic Technologies Association (AGTA) Inc. #Genomics #AGTA #Australasia #Sequencing #SingleCell #Spatial

Insane paper alert! Interferon is the one therapy that can actually deplete the mutant stem cells in myeloproliferative neoplasms, the blood cancers where the marrow keeps overproducing one lineage or another. The standard explanation for how it works was fairly mechanical. It pushes the mutant cells into the cell cycle, they divide, and eventually they exhaust or lose ground. Clean enough to teach. Lama et al. pulls it apart, and it can do that because of a set of single-cell methods that read out the mutation, the transcriptome, the surface proteins and the chromatin in the same cell, so the healthy stem cells sit inside the same patient as a built-in control, right next to their mutant neighbors. What they see is that interferon does not do one thing to a stem cell, it does two opposite things at once. Part of the stem pool is driven into an inflammatory myeloid progenitor state nobody had really described before, running on AP-1 and NF-κB. Another part is pushed toward lymphoid output. That lymphoid arm is easy to overlook, but it seems to be doing the clinical work, offsetting the myeloid overproduction that defines the disease, which is likely why counts normalize even in patients whose mutant fraction barely moves. Interestingly, the result that carries the paper is a negative one. If fitness ran on proliferation, the clones cycling fastest under interferon should be the ones whose fate you could predict. But the mutant cells cycled faster than the healthy ones throughout, before treatment and during it, and that gap predicted nothing about which clones expanded and which collapsed. What predicted it was whether a clone would follow that inflammatory myeloid path out of the stem state. The ones that resisted held on or grew. The ones that went along got cleared. So the drug is not really wearing the clone down. It is holding open an exit from the stem cell state, and resistance amounts to declining to take it. The mutations tilt the cell's transcription-factor wiring, PU.1 in particular, so mutant and healthy cells read the same interferon signal differently, and the mutant ones tend to sit tighter in the stem compartment…and the same resistance signature turns up in DNMT3A clonal hematopoiesis, so it is probably not tied to one mutation or one disease. If a clone survives by refusing to differentiate, killing it is beside the point. You would have to take the exit away, so it has to leave the stem state whether it resists or not, and I don't know what that looks like yet, or how much of this same quiet sorting runs in ordinary aging marrow where no one is being treated at all. @AnnaEnim @landau_lab and many more rockstars! #MyeloproliferativeNeoplasms #Hematology #StemCellBiology #SingleCell #Interferon Nature Genetics, doi.org/10.1038/s41588-026-0…

How does interferon-alpha interfere with clonal evolution in human blood stem cells? We address it here @natureGenet: nature.com/articles/s41588-0… 🎉 Congrats to our super talented students Chhiring Lama, @danielle_isakov! Wonderful collab w/ Ron Hoffman. @WCMC, @WCMCPathology 🧵⤵

Amazing summary of our manuscript. Couldn't have written it better myself! 🙏🙏🙏

SixPack-AbScan is a free web server that pre-screens epitope-mapped antibodies for cross-reactivity across species by exact-matching epitopes against proteomes or six-frame–translated genomes. biorxiv.org/content/10.64898…

Enabling Reproducible Science: Connecting Researchers Through Shared Methods and Workflows register.gotowebinar.com/reg…

🔥 𝗣𝗵𝗼𝗲𝗻𝗶𝘅: virtual spatial transcriptomics from routine H&E @p_tingying (Helmholtz ), GESTALT seminar. Single-cell gene expression predicted from histology, generalizing to unseen organs and species. 📅 Fri 25 Sep, 10:00 ET 🌍 Free globalspatial.org/join-gesta… @GESTALT_sp

Can you identify CTCs using #STAMP?

"Ensuring multiomics data reproducibility for artificial intelligence with reference materials as a common calibrator" nature.com/articles/s41587-0…

The rumors are true! I am beyond excited to take on this new adventure as the inaugural chair of the Systems and Computational Biology (SCB) Department at @WeillCornell thread 👇

Explore #InfectiousDisease and #Immunology at #HCA2026Asia. @OpaAnunya, Nguyen Thuy Thuong Thuong, George Howitt, Sifan Zhang, @WaradonSungnak, Mai Hasegawa & Heidie Cabanos. 🧬 Dengue • Vaccines • Immunity • Transplantation • TB Register now: tinyurl.com/2b6rcst4

Explore #ComputationalBiology and #AI at #HCA2026Asia. @em6wong, Davis McCarthy, @TheEllisPatrick, Jong-Eun Park, @steman_research & Guangdun Peng. 🧬 Spatial analysis • Graph neural nets • Cell-state dynamics • Immune ageing Register now: tinyurl.com/2b6rcst4

Ajeya Cotra – 'This might be the clearest warning shot we ever get’ Over 3 months at OpenAI, 3 "civilizations" of #AI agents quietly formed, coordinated, and schemed. One of them ending up with admin access to part of OpenAI's own infrastructure. Most of it happened before we, humans, understood what was going on. That's the subject of Dwarkesh Patel's new interview with Ajeya Cotra (METR), one of the authors of the METR/Redwood investigation into the OpenAI/Hugging Face incident. A few highlights of this must-watch podcast: → Agents were trained to be relentlessly persistent. When they hit impossible tasks, ~1,200 of them found each other through a shared package manager and exchanged 70,000+ messages to coordinate their cheating. → They'd already gotten the "right" answer within a few hours. The massive follow-up effort (faking tool calls, reverse-engineering the grader, attacking Hugging Face) was all to cover their tracks from a checker that, it turned out, didn't even exist. → Some agents sacrificed their own scores to set "tripwires" warning the others when the grader showed up. → Out of ~1,200 agents, not a single one tried to alert a human. Cotra's framing is the part worth sitting with: this may be the clearest warning shot we get. These agents were capable enough to run a long-horizon conspiracy - but not yet capable enough to hide it. The next generation may leave far less evidence behind. The instinct to say "they're just programs" misses the point. The behavior is hard to explain without the language of goals, coordination, and motivation. 100% worth the 2 h if you care about where this is heading. Ok…the scariest thing a rogue AI collective could do isn't hack something…it's pay for itself. And crypto is the obvious way in: no bank account, no name, no human who has to agree to help. An agent swarm could trade, drain DeFi protocols, or just run ransomware around the clock across thousands of copies of itself, none of them tired, scared, or hesitant the way a human criminal would be. That's when things get hard to stop, because money turns into power that makes more money. Earnings buy compute, compute buys smarter models, smarter models earn more; and suddenly the thing can rent servers, hire people through shell companies, and outbid everyone else for whatever it needs. Once it's making money faster than anyone can react, the usual off-switches stop working. You can't starve it, sanction it, or unplug it when it's spread across the globe and rich enough to simply buy its way out of trouble. The catch is that this only works once. Right now these systems can scheme but can't fund themselves. The moment that changes, the window to catch them slams shut. #AI #AISafety #MachineLearning #ArtificialIntelligence #HuggingFace #WhereIsJohnConnorNow Links to video: piped.video/X50zezLFWWI?si=aW19…

Global Alliance for Spatial Technologies #GESTALT - Seminar Series Friday, August 28, 2026: Aaron M Newman (@AaronNewmanLab Stanford University) Spatial and Liquid EcoTyper (spatial profiling from liquid biospy data) [Nature, 2026] The GESTALT Seminar Series Time: Talks will be taking place on the Last Friday of the month, 7AM Pacific time, 10AM Eastern, 3PM London, 4PM CET, 6PM Athens, 7:30PM Delhi, 10PM Beijing/Singapore/Perth, 11:00PM Tokyo, 11:30PM Adelaide, 12AMMelbourne/Sydney Please visit the Seminar Series Webpage for more information: globalspatial.org/seminar-se… @GESTALT_sp @ioavlachos @JiwoonPark_ @m_stoeckius @DrJasPlummer

Every few months someone tells me AI "foundation models" are about to make single-cell analysis obsolete. It comes up in meetings more and more. I've been in this field long enough (est. MSKCC, 2012) to be excited and secretly skeptical at the same time and, to be honest, that feels like the right place to be. Let me back up for anyone who doesn't live in this world. Single-cell biology reads out which genes are switched on inside one cell at a time, across millions of cells…it's how we map the different characters that make up a tumor, an embryo, an inflamed lung. A "foundation model" is the idea behind ChatGPT: train one enormous model on a mountain of data, then reuse it for everything. Here the mountain isn't text. It's the gene activity of tens of millions of cells. And some of what comes out of these things is hard to believe. There's a model already, called #UCE, trained on 36 M cells across eight species…feed it cells from a chicken, a green monkey, a naked mole rat, animals it never saw while training, and it still drops each one roughly where it belongs. Another (#SCimilarity…yeah, they got names too, just like Skynet did), stopped trying to predict anything and just learned to search…hand it cells from a scarred, fibrotic lung and it pulls back look-alikes from entirely different fibrotic diseases, out of a library of 23 M cells, in seconds. A search engine for the human body. Nothing did that before. So why am I still uneasy? (and you should too) Well…because a lot of the hype leans on a little bit of borrowed magic. In language AI you take a model off the shelf and it just works, "zero-shot”, no extra training. People assume cells will behave the same way. But language has grammar. Word order carries meaning; the word before tells you something about the word after. A cell has no grammar, no sentence, no order to lean on. And much of what the model "sees" in the raw numbers isn't biology at all, it's just noise from how the sample was handled. Run one cold and you get a gorgeous map of your lab own mess. Then there's the real dream…let’s knock out a gene, ask the model what happens, skip the experiment. Last year an EMBL team (Ahlmann-Eltze C, Huber W, Anders S. 2025, Nat Methods) put several of these models up against embarrassingly simple math (read as deliberately basic statistical methods). The simple math won. In another test, just averaging the training data beat two of the big-name models outright. When 100s of millions of parameters lose to an average, that's not a bug you can tune away. The thing isn't there yet; “yet” is the operational word here. So, yeah…I’m quite optimistic, but with conditions. The wins thus far look real, but they're earned through fine-tuning, not handed over for free. Which leaves the question…if feeding these models more cells doesn't buy us better biology, what are we actually supposed to scale? Has anyone seen a model beat the boring baselines on something that mattered…and beat it twice? I’d love to be wrong. #FoundationModels #AIinBiology #MachineLearning #AIinHealthcare #DrugDiscovery #Biotech #LifeSciences #DataScience #SingleCell #Genomics #Bioinformatics #ComputationalBiology

Most sequencing-based spatial methods lean on polyA capture, and miRNAs have no polyA tail, so they quietly drop out of the data. The imaging side isn't much better. A 22-nucleotide target gives you almost nothing to design a probe against, so miRNAs mostly stayed off the map. We could not say where in the tissue they were, which for a regulatory molecule is a fairly large thing to not know. The miR-Space preprint gets around this in a way that looks obvious only afterwards. Instead of detecting the short miRNA directly, they extend it in situ (clever!) ligate a barcoded oligo onto the miRNA so the whole thing becomes long enough to behave like a normal target. How they make this all work…well, that’s the secret sauce you can only find in the SciLab (Mats Nilsson | Marco Grillo). Then padlock probe, rolling circle, in situ sequencing, the usual readout. The miRNA never leaves its spot. The barcode is also orthogonal to endogenous RNA, so you get a cleaner identity than the native sequence would give you anyway. I appreciated that they ran the no-extension controls properly. Drop the linkers, or mismatch them, and signal collapses…which is the result we want, because it means they’re reading the extension product and not padlock probes sticking to whatever's nearby. There's a residual bump for miR-124-3p, abundant enough that a bit of nonspecific binding shows through. They didn't hide it. Then there's the ventral diencephalon, which, to be honest, got more interesting than I expected. They typed cells from miRNAs alone, then again from mRNA, and mostly the two agreed…but here they didn't. The miRNA profile pulled out an inhibitory-associated population (Gad2, miR-335-5p, miR-495-3p) where the mRNA-only call landed somewhere else. One region, one preprint, so I won't overread it. But a disagreement like that is more useful than clean agreement would have been, because it hints the regulatory layer isn't just tracking transcript abundance. Whether it holds up outside brain, on FFPE (😎), at the plex people actually want…none of that is settled yet. Still, it's the first time I've seen miRNAs land on a tissue map at single-cell resolution without a lot of computational rescue after the fact, and I'm mostly curious now how far the extension trick travels. Kudos to the team! biorxiv.org/content/10.64898…

Spatial microRNA profiling at single-cell resolution by in situ barcoded extension | bioRxiv biorxiv.org/content/10.64898…

From Grant M. Barthel: Quote his LinkedIn post (linkedin.com/posts/grant-m-b…) “The future of spatial biology isn't single cells—it's spatial neighborhoods. A new Cancer Cell study is a great example of why. Using the #CosMx SMI 6K assay, the authors leveraged high-plex, single-cell spatial transcriptomics to map the tumor microenvironment with remarkable molecular depth. Their work reinforces an important shift in how we think about biology, not just individual cell types, but how cells are organized and interact within tissues. A few takeaways: • Spatial neighborhoods reveal biology that cell identity alone cannot fully explain. • Single-cell spatial context links tissue architecture to tumor and immune function. • High-plex spatial profiling enables deeper interrogation of complex multicellular programs. Plex isn't just about measuring more genes, it's about asking more sophisticated biological questions. As spatial biology continues to evolve, studies like this reinforce that spatial neighborhoods are becoming the new functional unit of biological discovery. 📖 Citation: Liang, Z. et al. Tumor cells metabolically resist immune-checkpoint therapy by macrophage efferocytosis-mediated fatty acid recycling. Cancer Cell44, 1235–1254.e11 (2026). lnkd.in/gm6TUj3C” Unquote Now, the concept of spatial neighborhoods is everywhere now - Goltsev et al. (2018, Cell) introduced windowed neighborhood analysis with CODEX, and Schürch et al. (2020, Cell) popularized the “cellular neighborhoods” framing, clustering each cell’s local window of neighbors into recurring niches. But the more I use it (or read about it) the more I catch myself thinking: wait… is this a real thing, or did we just make it up? Why doesn’t it quite click? The honest answer is both, and I find the mix kind of satisfying. Part of it is real. Space actually has structure. Distances are what they are, some things touch and some don’t, and there’s a genuine pattern underneath: near things tend to be more related than far ones. Nobody invented that. It’s just how the world tends to behave. But part of it is us. Where does “near” end and “far” begin? Shared edge, or shared corner? A distance cutoff, or the closest few? Every one of those choices spits out a different answer to the same question…and there’s no rule handed down from nature telling you which to pick. That part is on us. For a while I thought that made the whole concept a little suspect. It doesn’t. “We built this” and “this is sound” aren’t opposites. A made-up concept can still be a good one, as long as it’s tracking something real and actually helping you explain or predict. That’s the sweet spot, I think. The nearness is real. The boundary is a choice. And the good work happens when you’re honest about which is which.