Bits In Bio Weekly - May 14th 2025
Updates From and Around the Bits in Bio Community
📢 Highlights
UK’s NHS trains sovereign GPT supermodel on 57 M Brit health records
Nabla’s “test-time scaling” makes a 1-B-param model hit heavyweight scores for protein AI
FDA Regulator rewires legacy tech stack with inaugural AI chief and Agency-wide infrastructure rollout planned for June.
Not yet a member of our super awesome slack community of >9000? Join HERE 🤗
👀 In Case You Missed it …
UK’s Foresight AI Initiative mines 10 B events to forecast population health
A new NHS England case study details Foresight AI, a Llama-2 based large-language model trained inside their ‘Secure Data Environment’ (SDW) on de-identified records from 57 M Brits—over 10 B healthcare events across 40 K codes. NHS England Digital Built by UCL and KCL with AWS/Databricks compute, the model autocompletes medical timelines, predicting admissions, diagnoses, and treatments post-COVID. By staying within the air tight secure data environment, data never leave NHS firewalls, addressing privacy critics while enabling rare-disease representation. Early pilots suggest Foresight can aid resource planning and shift care from treatment to prevention, yet full deployment awaits validation against clinician workflows. The project hints at how national health systems could build sovereign, population-scale GPTs without sharing raw data with Big Tech.
Nabla Bio debuts test-time scaling trick for de-novo antibody design
Cambridge based Nabla Bio has released a technical report showing its generative antibody model can be “super-sized” after training by sampling more parameters at inference—dubbed test-time scaling.The approach lets a 1-billion-parameter backbone punch above its weight, improving docking scores and developability metrics without retraining on new data. Early benchmarks suggest the scaled model rivals state-of-the-art structure-guided systems on GPCR epitopes—traditionally antibody-refractory targets. If validated, the trick could cut training costs while keeping design agility, an enticing combo for partners eyeing bespoke antibodies. Nabla hints at integrating the method into its wet-lab loop later this year, setting the stage for faster lead-to-candidate cycles.
FDA Embraces AI with Appointment of First Chief AI Officer and New Infrastructure for Accelerated Scientific Reviews
The FDA has named Jeremy Walsh as its first Chief AI Officer, has announced its integration strategy for their first generative AI system, marking a significant step toward modernizing its legacy infrastructure and processes. Walsh brings 14 years of experience in healthcare AI to the role and is taking helm of a multi year initiative in the making rather than launching it fresh under the current administration. According to FDA Commissioner Martin Makary, the agency plans to deploy a generative AI system to streamline scientific reviews and reduce the workload on FDA scientists. The system is expected to be integrated into the FDA’s internal platforms by June 30. However, concerns remain as proposed budget cuts under the Trump administration and a lack of clear regulatory strategy for AI have raised questions about the FDA’s capacity to keep pace with rapidly advancing medical technologies, a standing critique that has lead to the launch of the Office of Research Innovation, Validation, and Application (ORIVA) just earlier this month.
Intrepid Labs Raises $7M to Unite Drug Formulation with AI and Robotics
Intrepid Labs, a University of Toronto spinout, has emerged from stealth with serie A of $7 million in funding to commercialize Valiant, its autonomous drug formulation platform. Valiant combines machine learning and on-premises robotic lab automation to accelerate and optimize the complex process of drug formulation and compounding—a critical step in turning chemical isolates into effective and stable medicines by fine-tuning variables like drug load, polymer ratios, powder grain distribution, and solvents. Traditional formulation workflows rely heavily on single-variable testing, often leaving an alleged 99.9% of potential combinations unexplored. Valiant uses an active learning loop: an AI engine generates formulations, a robotic system tests them, and the results are fed back to update the model, which guides the next set of experiments. The platform can optimize across multiple objectives including manufacturability and absorption, and has demonstrated a 4–11× reduction in experimental runs. For example, formulations that typically require 85–427 experiments can now be achieved in just 18–88 runs. Another experiment aiming to improve oral absorption revealed the system was able to optimize the formulation in a week, achieving a 3-fold increase from systemic exposure compared to the original formulation. Intrepid has already secured commercial agreements with major pharmaceutical companies as it brings this next-generation R&D platform to market. The process of compounding, however, is a comparatively deterministic process compared to target identification and drug development.
Trump Administration Takes Aim at Drug Prices
In a Financial Times interview, Donald Trump vowed to slash U.S. drug prices by “up to 80 %,” reviving talk of an executive-order style “most-favoured-nation” policy. The promise lands as CMS readies its third round of Inflation Reduction Act negotiations and as PBMs already face pricing pressure. While few believe an 80 % haircut is politically or logistically feasible, the rhetoric alone can chill biotech deal-making and hurt orphan-drug valuation assumptions. Large-cap pharma stocks dipped intraday after the remarks, a reminder that election-year policy noise moves markets even before laws change. Expect stepped-up lobbying and renewed scrutiny of list-price vs. net-price dynamics as the campaign season heats up.
Virtual-patient startup Valinor exits stealth with multi-omics playbook
Salt-Lake-City–based Valinor Discovery has emerged from stealth with an ambitious plan to simulate therapeutic response in “virtual patients” built from longitudinal, matched multi-omics and clinical-assay data. Instead of training on cell-line surrogates, Valinor is generating proprietary datasets that pair primary-cell readouts with patient biopsy and blood metrics—feeding generative models that forecast transcriptomic shifts, protein abundance, methylation changes, and clinical endpoints before any human is dosed. Early collaborators include Stanford’s Montgomery Lab and Helmholtz Munich’s Computational Health Center, which will co-develop open benchmarks to gauge clinical translatability of perturbation-prediction models. A partnership with Latch Bio provides a GxP-ready web portal so pharma clients can run in-silico dose–response experiments and automate downstream workflows. With advisors drawn from Genentech, J&J, and flagship academic centers, Valinor joins a crowded field betting that richer multimodal data—not bigger language models alone—will finally make virtual trials more than hype.
OpenAI evaluates LLMs for health related discussions
OpenAI has just released HealthBench, a new benchmark built with input from over 260 physicians across 60 countries to evaluate AI in real-world healthcare conversations. It includes 5,000 multi-turn cases in multiple languages and specialties, with expert-written rubrics to assess performance. The goal is to ground model evaluation in clinical judgment and better reflect how AI might support both patients and providers. Initial results show progress, but also highlight how far current models are from clinical readiness.
Seqera lands $26 M Series B To Expand Open-Source Bioinformatics Stack
Workflow-engine pioneer Seqera (creator of Nextflow and nf-core) closed a $26 M Series B led by Addition, with Speedinvest, Talis, Amino Collective and others participating. The Barcelona-born company aims to turn its once-academic project into a full “software-first platform” spanning pipelines, container orchestration, and AI copilots inside VS Code. Management argues that reproducible compute is now table-stakes for petabyte-scale omics and ML workloads—as evidenced by Recursion, Xaira, and Isomorphic Labs building in-house stacks. Funds will expand enterprise features (GxP compliance, hybrid-cloud bursting) and grow community programs around Nextflow. Count Seqera among the critical plumbing plays making techbio’s data deluge actually runnable.
Insitro cuts staff to extend cash runway
Embattled Bay Area bio AI company Insitro is laying off ~65 employees—22 % of its team—to stretch cash into 2027 while prioritizing metabolic-disease and neuro pipelines. CEO Daphne Koller cited “tumultuous market conditions,” echoing recent downsizes at 10x, Recursion, and Tempus-adjacent Pathos. The move comes seven months after Insitro inked three Eli Lilly deals around GalNAc-siRNA and antibodies, signaling that partnerships alone won’t offset R&D burn. Management still targets first INDs in 2026, betting that leaner ops plus proprietary multimodal data will get programs clinic-ready. Investors will watch whether the reset foreshadows more consolidation in the crowded AI-drug-discovery cohort.
Benchling deal with Modern to unify AI-ready research data
Moderna will roll its broader research organization—hundreds of scientists—onto Benchling’s cloud LIMS/ELN platform to streamline sample tracking and hook in-house ML models into one data spine. The move follows February layoffs in Moderna’s digital group (10% of staff) and signals a “buy vs. build” shift for lab informatics. Benchling’s dev kit (REST, Python SDK, webhooks) will integrate instrument feeds and a PostgreSQL warehouse so AI teams can mine experiments in real time. For Benchling, the win adds a top-five mRNA player to its 1,300-company roster—useful leverage as it eyes a rumored 2026 IPO. Expect rivals like Dotmatics, Scispot, and Emerald Cloud to tout their own AI-centric upgrades to stay competitive.
Unity’s UBX1325 hits non-inferiority in DME but company moves to liquidate
Anti-aging company Unity Biotechnology’s BCL-xL inhibitor UBX1325 delivered vision gains statistically non-inferior to aflibercept through 36 weeks in tough-to-treat diabetic macular-edema patients. Sub-analysis showed better results in eyes with baseline CST < 400 µm, hinting at a viable niche. Nonetheless, Unity’s board approved a plan to slash burn, lay off all remaining staff (including its CEO), and explore M&A or asset sales. The pivot underscores the senolytics sector’s capital crunch as investors want later-stage derisking and tangible cures for real diseases. Eyes now turn to whether a larger ophthalmology player will license UBX1325 or if the program stalls like prior Unity shots on goal. Backed by SF natives and the like of Foudners Fund with 9 figures of financine and valuation once over 700M, Unity may presage Brian Armstrong’s move into the longevity bio space.
Insilico Medicine files third Hong Kong IPO in push for public capital
Third time’s the charm? Fresh off a $110 M Series E, Insilico Medicine re-submitted an IPO prospectus to HKEX—its third try since September 2024. The filing discloses 2023 R&D spend of $97 M and a 9-drug clinical pipeline led by rentosertib (TNIK, Ph 2 IPF). Debt financing from HSBC last year and minority stakes held by Warburg Pincus-backed Mesolite set the pre-IPO cap table. Management hopes renewed investor appetite for “AI-native” biotechs (Duality Bio IPO’d for $194 M) will offset macro jitters and U.S. regulatory drag. Success would give the sector its first large AI-drug float since Exscientia 2021—and a new valuation benchmark after the Recursion-Exscientia merger reset comps. Long looked at with a concern of awe and skepticim, In-Silico’s inability to raise at favorable terms in private markets is curious.
🗓️ Upcoming BiB Events
See our handy dandy Lu.ma event calendar HERE, please RSVP so folks can plan accordingly!
DataDrivenPharma Evening Forum at Bakar Labs, Wednesday, May 14
5:00 PM @ Woo Hon Fai Hall in Berkely
Chips & Cells: Advancing discoveries in cell membrane biology through AI & physics-based computational methods Friday May 16th in Berlin
Chicago Bits in Bio Happy Hour! Thursday, May 29 at Recess in Chicago
Personalized Medicine: The Role of Biotech in Tailoring Healthcare, Thursday May 29th in Lisbon
📰 Top Community Conversations
Not much to report this week - have a question?
Post it in our #ask-bib slack channel
🏢 New Job Openings
Director, Computer-Aided Drug Design/Computational Chemistry at Oddity Labs
UI/UX Designer at QBench
Bioinformatics Scientist at The Align Foundation
Director of Protein Projects at The Align Foundation
Senior Software Developer at Prenosis
Cheminformatician at Kiin AI
🙋 🙏 Community Asks
Feedback: How is the Newsletter doing? We’re trying different formats/content. In case the hyperlink above didn’t get your attention, maybe a bright orange button will!
Volunteer: Want to get involved with Bits in Bio, meet new members across the community, and learn about the ecosystem? We are looking for volunteers to help us create great content and manage the community.
🙏 Thank you for being a BiB Weekly reader!
We want to deliver what matters most in Bio AI and would love your feedback on how we can do better. Please weigh in as anon here or DM me directly!