Genomic Research Uses

Nine in ten drug programs that enter the clinic fail, most often when we first test for efficacy - when the mechanism did not drive the disease. The best-validated fix has been known for a decade: targets with human genetic support are 2 to 4 times more likely to succeed in the clinic. Yet only 3.6% of genetically supported targets have ever been pursued for an indication the genetics supports. The genetic window is foggy. A disease map yields dozens or hundreds of associated genes, most with muted effects, because evolution selects against variants of massive impact. Part 2 of my Deep Phenotype manifesto, out today, describes the AI we built at insitro to cut through that fog. Specialized ML models turn high-content measurements, from whole-body imaging in large human cohorts to genome-scale perturbation screens in human cells, into a library of precision phenotypes: quantitative traits that sharply raise the power of human genetics. In MASH, they surfaced over 30x more genetic associations than clinical staging. On top of this library sits a causal AI model, our Virtual Human, that integrates evidence across data modalities, disease biologies, and physical scales. The premise is that this integration gives rise to higher conviction, which in turn we believe will give rise to higher clinical success rates. To assess that hypothesis,we asked the model to predict success of historical phase 2 trials, zero-shot, with no trial outcomes in training. In its top decile of target-indication pairs, the false-positive rate was 10% in metabolic disease and 21% in cardiac, against a historical failure rate of 57% in both. We recognize, of course, that a retrospective benchmark is no guarantee of prospective success. Conviction is only half the value. Because the same phenotypes span from patient to cell, a genetic hit becomes an experiment we can run: the platform tells us in which cells a gene acts, which pathways it perturbs, whether a drug should inhibit or activate it. The assays that credential a target become the assays we optimize molecules against; the biomarkers that found the mechanism follow the drug into the clinic. The piece traces one example end to end: MASH, from the UK Biobank to a validated liver program. If this works, the payoff arrives twice. The first is economic: every medicine that reaches a patient carries the cost of the failures behind it, and nothing lowers that burden more than mechanisms that survive phase 2. The larger prize is measured in patients: the many diseases that still lack any disease-modifying therapy because we have not known which mechanisms drive them. If we can generate causally credentialed targets repeatably, disease after disease, the question changes: from whether the next program will fare better to how many diseases we can take on at once. That is the wager behind the Virtual Human: a repeatable path to medicines for the many who have none today. https://lnkd.in/gxFt-Fst

Half a million genomes. 1.5 billion variants. One breakthrough: we are all truly unique. Twenty years ago, the Human Genome Project took 13 years and $2.7B to sequence a single genome. Today? We can sequence a genome in less than 24 hours for under $1,000. Last week, UK Biobank released 490,640 whole genomes — the largest genetic dataset ever (Nature, 2025). What did we learn? • Each person carries 4–5 million variants • 76% appear in fewer than 10 people — your genome is almost entirely yours • 1 in 10 carries clinically actionable mutations where doctors can intervene today (e.g., BRCA1/2 for cancer, LDLR for heart disease) Why it matters: • Previous genetic tests captured ~6% of human variation. This dataset reveals 40× more • In non-coding regions — the biological switches controlling genes — researchers found 63 new disease associations • Adding 31,785 non-European genomes uncovered 82 disease links invisible in Eurocentric studies From genetics to health impact This transforms medicine today: • Prevention - Polygenic risk scores flag disease decades before symptoms • Diagnosis - Rare disease patients waiting years for answers finally find them • Treatment - Pharmacogenomics matches the right drug, right dose, to your genome The next frontier: genetics + everything else Genetics is the hardware. Health is the software running in real time. Your DNA is fixed, but biology is dynamic, shaped by: • Epigenetics: how environment and lifestyle switch genes on/off • Proteomics & metabolomics: molecular signals revealing your current health state • Digital biomarkers: continuous data from stress, sleep, glucose, heart rate • Stress biology & neuroendocrine signaling: how cortisol and brain-body responses reshape your health trajectory Layer these dynamic signals onto genetic foundations, power them with AI, and you create living health models, not just predicting disease, but understanding when, why, and how it manifests in YOU. The critical question? We've spent decades treating the "average patient" — who doesn't exist. Now we can better see each person as they truly are: biologically unique, dynamically changing, infinitely complex. The healthcare winners of the next decade won't just collect data: they'll integrate genetics, epigenetics, molecular and phenotypic tests, lifestyle, stress biology, and digital signals to deliver truly personalized, preventive care at scale. There is no "normal" genome, only 8 billion unique experiments in being human. And we just decoded the first half million. 👉 Which excites you more: knowing your genetic blueprint, or understanding how your daily choices rewrite it?

So what is it now — four bases, five, or six? For years, we thought DNA was simple: A, T, C, G. Then epigenetics complicated the picture. Cytosine could be methylated (5mC). Later we learned it could be hydroxymethylated (5hmC). Suddenly C wasn’t just C anymore. Think of it like language. The letters on the page don’t change — but punctuation can flip the meaning. “Lets eat Grandma” is very different from “Let’s eat, Grandma.” That’s methylation: the marks don’t rewrite DNA, they control how it’s read. This matters. Because in cancer, methylation changes often appear before mutations do. They are early warning signals that show up in circulating tumor DNA long before scans or symptoms. And the technology landscape is heating up: Illumina just launched a new library prep tool for methylation sequencing Biomodal has a chemistry that reads 5mC and 5hmC together in the same workflow PacBio and Oxford Nanopore promise direct methylation readout from native DNA without conversions or extra prep So who wins? The incumbent with scale and throughput? The innovator with richer signals? Or the platforms that bypass chemistry altogether with long native reads? This isn’t an academic debate. The answer will shape how we detect cancer, track relapse, and even monitor therapy response. My experience says the winner won’t just be whoever has the best science. It will be whoever makes methylation analysis routine, affordable, and robust at scale in messy real-world samples. Are we living in a five-base world, or is six the new normal? And more importantly — who gets there first? Curious where you’d place your bet. #Genomics #Epigenetics #CancerResearch #cfDNA #LiquidBiopsy #Diagnostics

cientists found that a new blood test analyzing cell-free DNA (cfDNA) in the bloodstream can detect many types of cancer early with impressive accuracy. This test used machine learning to study DNA fragments shed by tumors. In trials involving thousands of patients, the test showed an 87.4% sensitivity, meaning it correctly identified cancer cases, and a 97.8% specificity, meaning it rarely gave false positives. It could also predict the tissue where the cancer started about 83% of the time. The test worked well even for early-stage cancers like liver, lung, ovarian, and pancreatic cancers, which are usually hard to detect early. In a study with asymptomatic people, the test identified over half of cancers within a year, many of which standard screenings missed. These results suggest this cfDNA-based test could be a powerful new tool to catch cancer earlier and guide treatment decisions, potentially saving more lives by spotting cancers sooner than current methods allow.

CRISPR used to remove extra chromosomes in Down syndrome and restore cell function Japanese scientists report that it is possible to cut away the surplus chromosome in affected cells, which appears to bring their behavior closer to typical function. Scientists carefully design CRISPR guides to target only the unwanted chromosome. This trick is called allele-specific editing, and it helps steer the cutting enzyme to the right spot. Their group discovered that removing the unneeded copy often normalized gene expression in laboratory-grown cells. The treated cells reverted to typical patterns of protein manufacturing. They also showed better survival rates in certain tests, indicating that the excess genetic burden was successfully relieved. The researchers didn’t just test their approach on lab-grown stem cells. They also applied it to skin fibroblasts, which are more mature, non-stem cells taken from people with Down syndrome. The project shows that CRISPR can remove an entire chromosome rather than making small fixes. That is a big jump in what genome editing can accomplish. The study is published in PNAS Nexus. Source in comments.

This isn't just an ordinary test tube! 🧪 It contains genetically modified E. coli bacteria that glow bright green under UV light! 💚 because of "GFP protein" produced by these bacteria. 💡" This is a classic and beautiful demonstration of genetic engineering, commonly used in biology labs for research and education. 1. The Key Player: Green Fluorescent Protein (GFP) GFP was originally discovered not in bacteria, but in a jellyfish called Aequorea victoria. This jellyfish produces GFP naturally, which causes it to glow green in the deep ocean. · The Nobel Prize: The scientists who discovered and developed GFP, Osamu Shimomura, Martin Chalfie, and Roger Y. Tsien, were awarded the Nobel Prize in Chemistry in 2008. Their work revolutionized biology by allowing us to see processes inside living cells that were previously invisible. 2. How the Bacteria are Made to Glow: *Isolation: The gene (a specific piece of DNA) that contains the instructions for making the GFP protein is isolated from the jellyfish's DNA. *Insertion: This GFP gene is then inserted into a small, circular piece of DNA called a plasmid. Plasmids are like molecular taxis that can carry new genes into a cell. *Transformation: The engineered plasmid is introduced into the E. coli bacteria in a process called transformation. The bacteria then start reading the new instructions on the plasmid. *Production and Glow: The bacteria's own cellular machinery (ribosomes) follows the GFP gene's instructions to assemble the GFP protein. Once produced, the protein folds into its unique shape. When this specific shape is exposed to ultraviolet (UV) or blue light, it absorbs the energy and re-emits it as a visible, brilliant green light. This process is called fluorescence. 3. Why Use E. coli? · Workhorse of the Lab: E. coli is a simple, single-celled organism that is very well-understood, easy to grow, and reproduces quickly. This makes it the perfect "factory" for producing proteins like GFP. · Safety: The specific strain of E. coli used in these experiments is harmless and cannot survive outside the lab environment. 4. Why is This So Important? (Beyond Looking Cool) The true power of GFP is its application as a"reporter gene" or a "tag." Scientists can attach the GFP gene to the gene of another protein they want to study. · For example, they can create a "glow-in-the-dark" version of insulin, a cancer-related protein, or a neuronal protein in the brain. · This allows them to visually track in real-time: · Where that protein goes inside a cell. · When it is produced. · How it moves and interacts with other molecules. In short, this test tube contains a tiny biological factory that produces one of the most important tools in modern molecular biology and medicine! It's a perfect example of how basic research on jellyfish can lead to groundbreaking technologies.

Today OpenMed released a dataset the likes of which I have never seen before. They just put over 1 billion rows of psychiatric genetics data on Hugging Face. ADHD. Depression. Schizophrenia. Bipolar disorder. PTSD. OCD. Autism. Anxiety. Tourette syndrome. Eating disorders. 12 conditions. 52 landmark studies. Every genome-wide association study (GWAS) ever published by the Psychiatric Genomics Consortium, but now standardized and accessible in one place. Previously, accessing this data meant tracking down dozens of files scattered across FTP servers, wrestling with inconsistent formats, and spending more time debugging download scripts than doing actual science. Now it's one line of Python! Each row represents a statistical test: how strongly is this specific point in the human genome associated with this psychiatric condition? In more depth, each row is a single variant-phenotype association test from a GWAS meta-analysis. For every SNP, you get: • Variant ID (e.g. rs6702460) and genomic location (CHR/POS) • Effect allele and reference allele (A1/A2) • Effect size — BETA or OR — and its standard error (SE) • P-value, imputation quality (INFO), allele frequency (FRQ/MAF) • Sample sizes — total, cases, and controls (N/Nca/Nco) A typical single GWAS tests 7–15 million variants. We have 52 of them, many with multiple ancestry groups and sub-analyses. That's how you get to 1.14 billion rows. Hopefully, this will mean: 🧠 Earlier identification of people at genetic risk for psychiatric conditions (even before symptoms emerge!) 💊 Better drug targets: pinpointing the genes and biological pathways causally involved across multiple disorders 🔗 Understanding why conditions co-occur (e.g., depression and anxiety, ADHD and autism) through shared genetic architecture 🌍 More equitable research: ancestry-stratified data means findings that apply beyond predominantly European study populations Mental health research has long been underpowered relative to the scale of the problem. Open, accessible data is one way to change that 🤗

Nature (5 Aug 26) published three coordinated papers that together generate and validate the largest clinically annotated collection of patient-derived 3D cancer models assembled to date, and demonstrate that these models reveal therapeutic gene dependencies undetectable in traditional 2D cell lines. 🔅 The Human Cancer Models Initiative (HCMI) generated 665 models from 637 patients across 25 cancer types, from a consent cohort of 2,780 donors, including 153 rare cancer models and 71 models from non-European donors. Across 421 matched tumour-model pairs, DNA concordance was 97.8% and epigenetic concordance 95%. Single-nucleus RNA sequencing identified three mechanisms of model-tumour divergence: stromal purification, clonal selection, and culture-medium-driven epigenomic plasticity, the last reversible by switching growth conditions. Extrachromosomal DNA was model-specific in 43.9% of cases, yet MYCN, KRAS and EGFR amplifications were preserved in a subset. GBM models from patients with prolonged temozolomide exposure retained the SBS11 treatment resistance mutational signature. All models are available via ATCC, the NCI GDC portal and the HCMI Explorer Suite. 🔅 The Wellcome Sanger Institute derived 256 clinically annotated organoids from five cancer types and completed genome-wide CRISPR-Cas9 screens across 162 organoids (AUROC = 0.97), identifying 97 core fitness genes unique to 3D organoid cultures, enriched in isoprenoid and steroid biosynthesis. In colorectal organoids, KRAS dependency varied by allele: G12X variants retained co-dependency on upstream EGFR signalling while Q61H organoids were fully unresponsive to EGFR inhibition, consistent with signalling-independent oncogenic activation. In paired pre- and post-treatment oesophageal organoids, chemotherapy-driven clonal evolution reduced KRAS and DNMT1 dependency and increased PSMB5 dependency, confirmed pharmacologically with proteasome inhibitors bortezomib and ixazomib. 🔅 The Broad Institute added 314 NextGen models across 10 cancer types to DepMap via 147 CRISPR screens. NextGen models matched annotated cancer lineage in 69% of cases versus 35% for traditional cell lines; CNS spheroids matched at 93% versus 11%. A PDAC-classical/mucinous transcriptional program, preserved in GI organoids but silenced in 2D lines, correlated with WNT pathway dependency (WLS, MESD, FZD5, LGR4 and TCF7L2). In glial GBM spheroids, CDKN2A loss predicted CDK6 dependency and CDK4/6 inhibitor sensitivity. Direct comparison of 3D versus 2D conditions showed that serum-containing media masked SCD dependency in KRAS-amplified oesophagus-stomach organoids, demonstrating that growth medium independently shapes gene essentiality. 📑 HCMI compendium: https://lnkd.in/gjgFTp-3 📑 Sanger organoid biobank: https://lnkd.in/gFXhqeii 📑 Broad DepMap NextGen: https://lnkd.in/ggbssEH4 #CancerResearch #PrecisionOncology #CRISPR

A mind-blowing paper from the UK Biobank last week: whole genome sequencing of 490,640 participants. To put this into context, the Human Genome Project built a single human genome reference by sequencing DNA from a handful of people. It cost ~$3B and took around 13 years. The latest work is nearly half a million people. Sequencing at a rate of ~115,000 genomes per year, and cost a few hundred dollars per genome. --- A big challenge in genetics is linking genetic variations to an outcome (disease). Variations can be single DNA base changes, additions or deletions of DNA, which are found in a percentage of the population In this latest paper, they identified ~1.5 billion variations. And each UK biobank participant provides extensive phenotypic data (eg health questionnaires, demographics, proteomics, metabolomics…). So this is a major step in linking genetic variation to diseases and improving diagnostics. I’m a big fan of this project, and there are some friends on the paper (congratulations). This project is a great win for UK and world science. p.s. I’ll add the paper to the comments.

I am tremendously excited about the real-world impact of our latest publication on #AI #Biomarkers in Nature Medicine: https://lnkd.in/dv-7aS7Y Even in the US barely half of #lungcancer patients are tested for #EGFR mutations, for which targeted therapies readily exist. We have worked for many, many years now to try to overcome this gap with AI for H&E slides to offer patients a fast and cost-effective solution to get the right treatment. The point of this work is not only that we actually built it, but that Gabriele Campanella and Chad Vanderbilt organized a consortium and created the infrastructure for the first real-world, real-time deployment of a fine-tuned pathology foundation model for lung cancer biomarker detection. 𝙋𝙧𝙤𝙨𝙥𝙚𝙘𝙩𝙞𝙫𝙚𝙡𝙮!   𝐌𝐞𝐞𝐭 𝐄𝐀𝐆𝐋𝐄 (EGFR AI Genomic Lung Evaluation): ✅ 𝟎.𝟖𝟗 𝐀𝐔𝐂 in a 𝐩𝐫𝐨𝐬𝐩𝐞𝐜𝐭𝐢𝐯𝐞 silent trial with clinical-grade performance. 🌍 Generalizes 𝐚𝐜𝐫𝐨𝐬𝐬 𝐡𝐨𝐬𝐩𝐢𝐭𝐚𝐥𝐬 𝐚𝐧𝐝 𝐜𝐨𝐧𝐭𝐢𝐧𝐞𝐧𝐭𝐬 with robustness and reproducibility. 🔬 Validated on 𝐢𝐧𝐭𝐞𝐫𝐧𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐜𝐨𝐡𝐨𝐫𝐭𝐬, 𝐦𝐮𝐥𝐭𝐢𝐩𝐥𝐞 𝐢𝐧𝐬𝐭𝐢𝐭𝐮𝐭𝐢𝐨𝐧𝐬, 𝐚𝐧𝐝 𝐬𝐜𝐚𝐧𝐧𝐞𝐫𝐬. 🧪 𝟒𝟑% 𝐫𝐞𝐝𝐮𝐜𝐭𝐢𝐨𝐧 𝐢𝐧 𝐫𝐚𝐩𝐢𝐝 𝐦𝐨𝐥𝐞𝐜𝐮𝐥𝐚𝐫 𝐭𝐞𝐬𝐭𝐬, preserving biopsy tissue for full genomic profiling. ⚡ 𝐃𝐞𝐥𝐢𝐯𝐞𝐫𝐬 𝐫𝐞𝐬𝐮𝐥𝐭𝐬 𝐢𝐧 𝐮𝐧𝐝𝐞𝐫 𝟏 𝐡𝐨𝐮𝐫, compared to 2–3 weeks for NGS. 🚀 A foundational step toward regulatory approval and 𝐀𝐈-𝐢𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐞𝐝 𝐜𝐥𝐢𝐧𝐢𝐜𝐚𝐥 𝐰𝐨𝐫𝐤𝐟𝐥𝐨𝐰𝐬.   We have worked on Computational Biomarkers in Pathology continuously for over a decade starting with AI for predicting SPOP in prostate cancer from H&E in 2015, but seeing everything come to fruition at such a scale in 2025 is very humbling. AI, when done right, can give real, tangible help to cancer patients. 𝑰𝒕 𝒊𝒔 𝒐𝒖𝒓 𝒓𝒆𝒔𝒑𝒐𝒏𝒔𝒊𝒃𝒊𝒍𝒊𝒕𝒚 𝒕𝒐 𝒎𝒂𝒌𝒆 𝒊𝒕 𝒂 𝒓𝒆𝒂𝒍𝒊𝒕𝒚! I am deeply grateful to everyone on this most amazing team: Gabriele Campanella, Neeraj Kumar, Ph.D., Swaraj Nanda, Siddharth Singi, Eugene Fluder, Ricky Kwan, Silke Mühlstedt, Nicole  Pfarr, Peter Schüffler, Ida Häggström, Noora Neittaanmäki, Levent Akyürek, Alina Basnet, Tamara Jamaspishvili, Michel Nasr, Matthew Croken, Fred Hirsch, Arielle Elkrief, Helena Yu, Orly Ardon, Greg Goldgof, Meera Hameed, Jane Houldsworth, Maria E. Arcila, Chad Vanderbilt #AI #ComputationalPathology #Biomarkers #AIinHealthcare #DigitalPathology #PrecisionMedicine #LungCancer #EGFR #NatureMedicine #FoundationModels #EAGLEModel #EAGLE #Oncology