Kian Sadeghi

Introducing Boltz-2, an open-source model from MIT and Recursion that predicts both protein structures and binding affinities. It reaches free-energy-perturbation (FEP) accuracy at ~1,000× the speed. Use it on Tamarind Bio today: https://lnkd.in/gqnAG8Tz Binding affinity determines whether a drug is potent enough; FEP is precise but slow, while docking is fast but noisy. Deep-learning alternatives have lagged, until now. Performance: • Matches OpenFE on the standard FEP+ benchmark while running 1,000× faster. • Tops the CASP16 affinity challenge (140 complexes). • In MF-PCBA hit-discovery screens, doubles average precision over ML and docking. Coupled with SynFlowNet, Boltz-2 powered a prospective TYK2 screen; ABFE predicted binding for all top-10 generated compounds. For structure prediction, Boltz-2 Matches or exceeds Boltz-1, with major improvements on DNA–protein, RNA, and antibody–antigen complexes. Specify contact constraints, templates, or experimental methods directly in the model. Architecture upgrades: Builds on Boltz-1 with a new affinity head, finer controllability, GPU optimizations, and extensive synthetic + MD training data. Boltz-2 brings FEP-level accuracy to deep learning, at a fraction of the compute cost, and slots cleanly into generative screening pipelines. Excited to see how Boltz-2 will be used for faster small molecule discovery! Code & model: https://lnkd.in/ezDA9pji Paper: https://bit.ly/boltz2-pdf

722

11 Comments

🧬 𝙎𝙞𝙣𝙜𝙡𝙚-𝘾𝙚𝙡𝙡 𝙍𝙉𝘼-𝙎𝙚𝙦 𝙫𝙨 𝘽𝙪𝙡𝙠 𝙍𝙉𝘼-𝙎𝙚𝙦: 𝙒𝙝𝙖𝙩’𝙨 𝙩𝙝𝙚 𝘿𝙞𝙛𝙛𝙚𝙧𝙚𝙣𝙘𝙚? When I first heard the term single-cell RNA-Seq, I thought — “Wait… isn’t that just RNA-Seq? Why the fancy name?” Turns out, it’s not just a fancy name. It’s a completely different way of looking at gene expression. Let me explain it the way I wish someone had explained it to me when I was starting out 👇 📌 Imagine… You have a smoothie. It’s made of strawberries, bananas, and mangoes. You take a sip — you know it’s sweet, but you can’t tell exactly how much banana or mango is in there. That’s 𝙗𝙪𝙡𝙠 𝙍𝙉𝘼-𝙎𝙚𝙦 — it blends all the cells together and gives you the average expression. It’s simple, affordable, and great for big-picture trends. Now instead, imagine tasting each fruit separately. You can tell exactly how sweet each mango is, how ripe the strawberries are, and even spot a sour one. That’s 𝙨𝙞𝙣𝙜𝙡𝙚-𝙘𝙚𝙡𝙡 𝙍𝙉𝘼-𝙎𝙚𝙦 — it looks at each cell individually. You see the rare cell types, the subtle differences, and the “hidden” subpopulations that bulk RNA-Seq would miss. 🔍 𝗪𝗵𝗲𝗻 𝘄𝗼𝘂𝗹𝗱 𝘆𝗼𝘂 𝘂𝘀𝗲 𝗲𝗮𝗰𝗵? 𝘽𝙪𝙡𝙠 𝙍𝙉𝘼-𝙎𝙚𝙦: Perfect if you want a broad overview, like comparing healthy vs diseased tissue. Think: “What’s the average change in gene expression between these groups?” 𝙎𝙞𝙣𝙜𝙡𝙚-𝘾𝙚𝙡𝙡 𝙍𝙉𝘼-𝙎𝙚𝙦: Ideal if you need to understand diversity within a sample — like identifying rare immune cells in a tumor. Think: “What exactly is each cell type doing?” 🛠 𝙃𝙤𝙬 𝙩𝙤 𝙀𝙭𝙥𝙡𝙤𝙧𝙚 𝙏𝙝𝙚𝙢 𝙔𝙤𝙪𝙧𝙨𝙚𝙡𝙛 (𝙉𝙤 𝙇𝙖𝙗 𝙉𝙚𝙚𝙙𝙚𝙙) For bulk RNA-Seq datasets, try: 🔗 GEO: https://lnkd.in/g_wDbDjp For single-cell datasets, explore: 🔗 10x Genomics Datasets: https://lnkd.in/giQd3cy6 🔗 GEO Single-Cell Portal: https://lnkd.in/gSB62XNK Even if you don’t analyze the raw files yet, just browsing the studies and reading how they were designed will make concepts click. 💡 𝙏𝙝𝙚 𝘽𝙞𝙜 𝙏𝙖𝙠𝙚𝙖𝙬𝙖𝙮 Bulk RNA-Seq is like zooming out to see the whole forest. Single-cell RNA-Seq is like zooming in to examine each tree. Both views are valuable — you just need to know which one fits your research question. 🎁 𝘽𝙤𝙣𝙪𝙨 𝙏𝙞𝙥 If you’re a beginner, start with bulk RNA-Seq. It’s easier to learn, has more tutorials, and builds your foundation. Then, when you’re ready for more complexity, move on to single-cell. 💬 𝙎𝙩𝙞𝙡𝙡 𝙐𝙣𝙨𝙪𝙧𝙚 𝙒𝙝𝙞𝙘𝙝 𝙩𝙤 𝘾𝙝𝙤𝙤𝙨𝙚? I help students and researchers design their RNA-Seq projects step-by-step — from dataset selection to workflow planning. If you want a quick roadmap, DM me "RNA-Seq Strategy" for a 1:1 session (₹799). Because the right choice now can save you weeks of frustration later 🚀 #RNASeq #SingleCellRNASeq #Bioinformatics #NGS #10xGenomics #GEO

140

5 Comments

An AI-ready lab notebook for life science doesn't need to be complex or expensive. At Pioneer Labs, we use a lab management system built entirely with generic software tools. It's simple, cheap, and easy to modify. Thanks so much to ElevenLabs for their original Notion ELN template!! 🫶🙏 We have gotten such good use out of it, and by writing this guide I hope to pay it forward! Check out the blog post: https://lnkd.in/euzGi8h8 for: - How we use it - How to use it yourself - Information on the corpus we've amassed - if you're an AI researcher and think you could get insights out of this, reach out!

135

4 Comments

Biotech startups, stop posting “Hiring for Director of AI.” Or any AI role, for that matter… The sudden influx of AI experts—many with no scientific background—often don’t grasp the complexities of biological data. They don’t understand: • How much of biology is still unknown. • Why you can’t just scrape public datasets and expect useful results. • The cost and difficulty of generating high-quality, problem-specific data. If you have a drug pipeline and some sort of platform, prioritize the drugs. Those are your real assets, the ones investors (usually) care about more. Your first data hire should be someone with a strong biology background and deep analytical expertise. They should also understand FAIR principles (which is rarer than you think). Think a well-rounded data scientist. Someone who can handle: ✔️ High-quality data preprocessing and QC ✔️ Omics data analysis and integration ✔️ Scalable and reproducible data pipelines ✔️ Standardized workflows and data governance You’ll save significant time/money and yourself from many headaches later on on if you start building your infrastructure and analytical pipelines the right way from the get-go. This is a lot for one person to do, so be realistic with timelines, hire a larger initial data team, or… Outsource to a full data team that can set all of this up the right way from the start and be with you for the long run [little plug for dovetail biopartners]. A solid internal data team or data science partner that intimately understands your science and what it takes to streamline decision making are huge assets, and they’ll be able to vet and build whatever AI solution you think is needed in the future. — I’m pretty adamant about this because I’ve seen so many people make the wrong hires… but I still what to hear your thoughts. 👇

265

33 Comments

A lot of people are missing the mark about what’s currently important for in-silico protein design. At the moment, the competition between different generative models is rather trite. All the models do is lower the generative search space. Sure, you need to choose task-specific models, but picking between different generative models right now is like choosing the best camera brand for a Vogue cover phooshoot. It’s going to help you take the picture, and you need the right lens, but if you don’t know the whole workflow, from lighting, to posing, editing, aesthetics, your camera is not going to get you that cover photo. Your experience and skill in that field will, though. Same thing in protein design. Right now, I really don’t care what architecture was used to design the proteins. Like you don’t care what brand of camera took a cover photo, what matters is the end-product. I care about how the protein designs perform when we rank and filter them through 15 *other* models for various things like liabilities, expression, structure; and then out of the few left, what are their absolute value predictions for things like Tm and other task-specific conditions. Next, can you make any small mutations to fix any issues on the good candidates - that’s the editing part. If you want to learn the tricks of the trade and how to rank and filter millions of candidates, by all means go for it! Because if all you’ve done is chosen a generative model that generates your molecule type, picked from a few thousand using that model’s confidence scores, then I promise that you will be disappointed. Today, the best thing you can do is generate millions of candidates from several different task-specific models and architectures. You will then spend most of your time building the models to rank and filter them for your specific conditions. If you aren’t set up for this already, clearly it would be very time consuming, a learning curve, and compute-intensive. Especially in science, where the nature is to push into new areas, I don’t see this analogy changing. Everyone on the planet has access to a crisp camera and lens on their phones. That hasn’t changed the fact that not everyone takes good photos nor can anyone shoot a full-length movie with it. Biology is complex. We can democratize the models all we want, but that’s not going to magically make every biologist and ML scientist proficient at creating in-silico protein design pipelines.

192

5 Comments

Today, Salt AI raised $10M at a $29M valuation to create an AI-enabled end-to-end workflow for protein design, docking and lab testing, Axios reports. Salt’s platform is designed to connect several stages of the drug development pipeline: ⚙️ Starting from protein structures in the Protein Data Bank. ⚙️ Using generative models such as RF-Diffusion to create new variants. ⚙️ Validating stability with AlphaFold, which the company says it has optimized for a 22× speedup. ⚙️ Filtering candidates for criteria like diversity and immunogenicity. ⚙️ Running docking simulations to assess binding potential. ⚙️ Passing the most promising designs into wet lab experiments. The approach is meant to shorten the gap between computational modeling and experimental validation, a known bottleneck in drug discovery. The company expects to begin raising a Series A round in early 2026. Follow Where Tech Meets Bio for weekly tech+bio roundups and topical deep dives. Image credit: Salt AI

251

28 Comments

Scientists at Weill Cornell Medicine have identified a key bacterial culprit—Clostridium perfringens type B—that produces a brain-targeting toxin called epsilon toxin (ETX), which may trigger or worsen multiple sclerosis. Found more frequently in the guts of MS patients, this bacterium releases ETX that can breach the blood–brain barrier and induce MS-like brain damage in animal models. This discovery strengthens the link between gut bacteria and neuroinflammation, suggesting that microbial toxins may initiate or accelerate MS in people with underlying susceptibility. The findings mark a potential shift in MS research and open up possibilities for novel microbiome-targeted therapies or early detection methods.

157

27 Comments

Every biotech breakthrough has one thing in common: It starts with organized data. Behind every CRISPR breakthrough or mRNA vaccine lies something unsexy but critical: well-structured data. Yet most labs still use paper notebooks alongside Excel spreadsheets. Legacy software crashes weekly. Lab instruments generate proprietary formats no one can easily analyze. This digital fragmentation isn't just annoying – it's killing innovation. Scientists spend 30-40% of their time just managing data, delaying discoveries by months or years. Experiments get needlessly repeated. Compliance audits trigger panic as teams scramble for documentation. AI is changing this equation. Modern systems extract structured data from unstructured sources with remarkable accuracy. NLP algorithms scan thousands of papers, finding connections humans miss. AI can now predict protein structures in hours instead of months, enabling researchers to tackle previously "undruggable" targets. Traditional drug development averages 10-12 years with costs exceeding $2B. AI adopters routinely shave years off these timelines. What's holding labs back? Cultural resistance: Researchers view standardized systems as constraints rather than enablers. Financial concerns: Modern platforms require significant investment. Security fears: Biotech data represents valuable intellectual property. Forward-thinking labs overcome these barriers with high-value pilots and hybrid approaches. At Scispot, we help biotech companies connect these dots – building IP while maintaining compliance through integrated data solutions. The gap between leaders and laggards widens daily. Tomorrow's biotech winners won't just have the best scientists – they'll have data infrastructure that lets those scientists focus on discovery instead of data wrangling.

76

6 Comments

Abbas Kazimi, CEO of computational chemistry innovator Nimbus Therapeutics, discusses how the company’s drug discovery approach has evolved with AI, its deal making successes, and how its business model provides liquidity to investors and while making it easier to structure deals. https://lnkd.in/dzRG2V3d

50

1 Comment

The Jerusalem Post, a prominent international newspaper, published an in-depth article about our company, Itay&Beyond. This article details our groundbreaking patient-derived brain-on-a-chip technology, which is revolutionizing drug development and testing for neurological and psychiatric disorders. With successful proof of concept (POC) studies on epilepsy and Autism Spectrum Disorders showing remarkable results, we are proud to introduce our technology to the world. We aim to lead the field of drug development for these disorders by harnessing innovative scientific approaches. We invite you to read the full article and follow us at itayandbeyond.com. #science #news #neurology #psychiatry #neuroscience #autism #autismspectrumdisorders #israel #innovation #startup #medicine #drugs #drugdevelopment #pharma #itayandbeyond #breakingnews #epilepsy #biology #research #future #futuristic https://lnkd.in/d_CnR3Gm

30

AstraZeneca just made a $555M AI pact with a San Francisco biotech - no clinical programs. No approved drugs. Just ONE thing that made every boardroom take notice: AstraZeneca has been steadily expanding its AI footprint. In 2025 alone, it announced a $5.3B AI deal with CSPC and a $200M oncology partnership with Tempus and Pathos AI. And now they just made one of their boldest platform bets yet. AstraZeneca just signed a $555M partnership with Algen Biotechnologies, a San Francisco biotech with zero approved products. What did they see? A platform so powerful it could generate dozens of drug targets continuously. Algen's platform is called AlgenBrain. It combines AI with CRISPR-based functional genomics to map how genes regulate immune pathways. These aren't one-off discoveries. AlgenBrain is a discovery engine that can continuously generate new targets for immune-mediated diseases. That means multiple shots on goal, not just a single product. For a partnership with no clinical data, that's rare. But this wasn't a product bet. It was a platform bet. Under the deal, Algen identifies and validates new immunology targets while AstraZeneca gains exclusive rights to develop therapies against them. Total value up to $555M in upfront and milestone payments. What's most striking is the stage where this partnership begins: before any drug programs exist. Target identification is one of the least predictable phases of R&D. Platforms like Algen's can shorten that timeline, lower attrition, and strengthen first-in-class potential across multiple indications. Milestone-based payments align both sides on long-term success while limiting AstraZeneca's risk. Pharma companies are moving upstream. They're partnering with AI platforms to build continuous discovery capacity rather than buying late-stage assets. The next wave of biopharma M&A won't chase blockbusters, it'll chase the systems that create them. After working on 90+ deals over 25 years, I'm watching this trend accelerate faster than most realize. 3 takeaways if you're building in biotech: • Platform over product. Scalable discovery commands premium value. • Structure for milestones. Balance reward and risk through staged terms. • Secure early alliances. First access often defines long-term differentiation. And if you're investing, this deal redefines value. It's not just about late-stage efficacy data, it's about solving upstream bottlenecks.

1,293

54 Comments

The clock is ticking on SBIR reauthorization. Unless Congress acts, the Small Business Innovation Research (SBIR) program will expire on September 30th. For Massachusetts’s biotech startups and the entire U.S. life sciences ecosystem, SBIR isn’t just another program—it’s a lifeline for early-stage innovation that can lead to breakthrough therapies for patients Over the past year, MassBio has been laser-focused on advocating for SBIR’s renewal: 🔷 We convened members with Senate Small Business Committee Chair Sen. Joni Ernst and Ranking Member Sen. Ed Markey to share real-world impacts. 🔷 We surveyed our member companies and used their feedback to shape formal testimony on the program’s importance. 🔷 We worked with Biotechnology Innovation Organization's Council of State Bioscience Associations (CSBA) to circulate a joint sign-on letter urging swift reauthorization. 🔷 Just last week, we were back in conversation with Chair Ernst’s office about the latest redraft of the INNOVATE Act - which would expand Strategic Breakthrough Awards, empowering NIH and other agencies to make transformative investments in capital-intensive fields like biotech and medtech.  🔷 And we’ll be in D.C. in two weeks to continue our advocacy. The U.S. cannot afford to let SBIR lapse for even one day. MassBio calls on Congress to come together to prioritize American innovation by delivering on SBIR reauthorization before the September 30 deadline. As valued MassBio members, please don’t hesitate to reach out to our policy team with questions (policy [at] massbio.org). We’re actively following debate and competing proposals and will provide updates as needed in this final stretch. And don’t forget to share your stories about what SBIR has meant to you, to your company, and to the patients you’re in the lab fighting for. #SBIR #INNOVATEAct #Bioeconomy #MassBio #Startups #Biotech

70

5 Comments

Prolific Machines is rethinking the way we manufacture biology by using light to control cells. At SynBioBeta, co-founder and CEO Deniz Kent, PhD introduced the company’s optogenetic manufacturing platform, which promises precise, low-cost control over cellular behavior. Much of our economy is either made from cells or could be, from therapeutic proteins to materials and beyond. But realizing that potential hinges on one thing: control. Prolific Machines’ system delivers that control through light—offering temporal precision that outperforms traditional methods. Instead of redesigning sequences or abandoning hard-to-make drugs, asset owners can now work with Prolific to develop optogenetic cell lines tailored to their production challenges. The result is the ability to toggle gene expression on demand, inside any cell, with software and custom-built hardware to support it. This isn’t hypothetical. The company has raised millions of dollars, built the largest-ever optogenetics demonstration in a pilot facility, and developed a platform that scales from microreactors to industrial stainless steel tanks. By integrating genetics, hardware, and a light-based programming layer, Prolific is enabling the next generation of biomanufacturing—fast, precise, and built for the complexity of modern biologics. If you can’t make it this year, lock in early bird pricing for next year and join the fun: https://lnkd.in/grUy5cEQ

185

4 Comments