Bits In Bio Weekly - June 10th 2025
Updates From and Around the Bits in Bio Community
📢 Highlights
Recursion & MIT unveil Boltz-2 – an open model for protein–ligand binding predictions at 1000x speed
Britain’s big science bet: £86B to boost innovation across the country
FutureHouse open-sources ether0: a Mistral 24B derived chemistry model that shows its work
Not yet a member of our super awesome slack community of >9500? Join HERE 🤗
👀 In Case You Missed it …
Recursion and MIT open-source Boltz-2, predicting molecular binding 1000× faster than physics methods
Utah Based techbio titan Recursion teamed up with MIT to launch Boltz-2, a next-gen AI model that predicts how tightly molecules will bind to proteins – at impressive speeds In tests, Boltz-2’s accuracy in estimating binding affinity approaches that of gold-standard physics simulations like FEP, but it runs up to 1,000 times faster. The model doesn’t just predict affinity; it co-folds protein–ligand 3D structures and even gauges dynamics (like B-factors) for a more realistic picture. Trained on Recursion’s massive proprietary dataset (including ~5 million experimental affinity measurements and molecular dynamics data) using their BioHive-2 supercomputer, Boltz-2 outperformed all entrants in the recent CASP16 binding affinity prediction challenge. And here’s the kicker for the community: Boltz-2 is open-sourced under the MIT license, with model weights and its whole training pipeline freely available. The hope is that drug hunters everywhere – academic or commercial – can leverage or tweak this model to virtually screen huge libraries for hits in a fraction of the time it used to take. It’s a striking example of the open-science trend in biotech, translating cutting-edge AI into a shared tool to accelerate discovery across the board. Available on github.
UK announces £86 billion science & tech investment plan
The UK government has unveiled a massive £86 billion package to turbocharge R&D in high-growth areas like tech and life sciences, as part of a “Modern Industrial Strategy” aimed at boosting the economy. By 2029/30, annual public R&D spending would hit £22.5B, up significantly, but what’s novel is how the money will be deployed: the plan will empower regional innovation clusters so places like Liverpool, Manchester, Belfast, and Cardiff can steer funding toward their local strengths. For example, Liverpool’s renowned life-sciences hub could get funds to accelerate drug discovery startups, while South Wales might channel investment into its semiconductor industry. A new £500M Local Innovation Fund will let local leaders decide R&D priorities, a decentralization meant to unlock talent outside the usual Golden Triangle (Oxford-Cambridge-London) bubble. The government is pitching this as a win-win: driving jobs and productivity in left-behind regions while solidifying the UK’s status as a science superpower. Ambitious as it is, the strategy faces the hurdle of execution – but if it succeeds, Britain’s next biotech unicorn might just emerge from a provincial lab with London-grade support.
FutureHouse open-sources ether0, chemistry reasoning model built on Mistral
SF-based AI startup FutureHouse has released ether0, a 24-billion-parameter chemistry-specialized language model designed to act like a diligent lab assistant. Built on the open Mistral 24B model and fine-tuned with reinforcement learning, ether0 works through problems step by step, generating drug-like molecular structures to satisfy complex prompts, e.g. “design a molecule with these functional groups that bypasses the blood-brain barrier”. Somewhat uniquely, the model’s answers include “reasoning tokens” – basically its chain-of-thought scribbles – which emerge during training as it tries to get the chemistry right. These can look odd (ether0 even invented a pseudo-terminology word “reductamol” that it kept using), but hint at the model’s attempt to balance equations and obey chemistry rules internally. The result is an AI that, while not necessarily great at casual conversation or standard benchmarks, can seemingly propose valid molecules that meet a spec – avoiding the common LLM pitfalls of mismatched atoms or impossible bonds. FutureHouse open-sourced ether0 under Apache 2.0, complete with model weights on Hugging Face, inviting the community to explore this early experiment in AI “scientific reasoning”.
Behind the Deal: Scorpion Spins Out Antares with $177M to Advance AI-Enabled Precision Oncology
Scorpion Therapeutics has launched Antares Therapeutics with a $177 million Series A, following the divestiture of its two clinical programs—STX‑478 to Eli Lilly for up to $2.5 billion, and EGFR inhibitors to Pierre Fabre. Antares, led by Scorpion co-founder Adam Friedman, retains Scorpion’s preclinical pipeline targeting previously undruggable cancer biology, with its most advanced program likely to be a small-molecule, precision oncology candidate expected to enter the clinic in 2026.
The new entity is backed by top-tier investors including Omega Funds, Atlas, Lightspeed, BVF, and Cormorant. The spinout reflects a growing trend: extract value from de-risked assets through strategic sales, then double down on early-stage innovation with a fresh capitalization structure. Antares is expected to incorporate AI platforms—such as those used for de novo molecular design, predictive pharmacology, and generative chemistry—to accelerate lead optimization and target expansion. These tools enable the efficient exploration of vast chemical spaces and enhance the probability of success for precision-targeted therapies. With proven medicinal chemistry expertise and next-gen AI tools, Antares is positioned to move fast, partner early, and remain capital-efficient while advancing a portfolio built for selective oncology innovation.
Kiin Bio raises €1.9M pre-seed to build an AI “Virtual Scientist” for rapid drug discovery
London-based Kiin Bio has raised a €1.9 million (≈$2.2M) pre-seed round to develop its so-called “Virtual Scientist” platform for drug discovery. The startup is allegedly combining synthetic biology and generative AI to simulate and screen potential new molecules with what it claims will be unprecedented speed and accuracy. In essence, they’re trying to create an AI-driven research system that can do a lot of the early trial-and-error of drug R&D in silico – rapidly generating chemical ideas, testing hypotheses via models, and narrowing down candidates before any wet-lab work begins. While the details as to how this approach is different from contemporaries with 100x their funding and 2-5 year lead, the round is led by notables b2venture with participation from firms like HEARTFELT_ and Rule 30. Kiin Bio joins a wave of young companies betting that smart algorithms (paired with real biological data) can act as “virtual scientists,” crunching through data and freeing human researchers to focus on the most promising leads. It’s very early-stage, but if successful, their platform could significantly cut down the time and cost needed to discover new drugs.
DeepSeek recruits medical interns to train its wildly popular AI for hospitals
One of the darlings of the open source AI movement, Chinese startup DeepSeek, fresh off seeing its eponymous large language models deployed in 300+ hospitals is taking direct steps to make them better: hiring interns with medical training to label clinical data. The company is offering about $70 a day (500 yuan) for med students or grads in Beijing who can spend four days a week curating and annotating cases for its “auxiliary diagnosis” tools. These interns aren’t your typical data labelers—they’re expected to know their way around Python and prompt engineering as well as patient charts, which is not a common cross section of skillsets. The goal is to reduce Deepseek AI’s tendency to provide subtly incorrect answers in high-stakes clinical situations. It’s the first time DeepSeek has openly called for “medical data” labeling, reflecting how seriously it’s tackling safety after a JAMA-published warning noted its model’s plausible-but-wrong outputs could pose “substantial clinical risk”. With a recent model update already reducing its error rate by 50%, DeepSeek’s crowdsourced strategy shows how hospitals’ new AI co-pilots might be domesticated—by teaming them up with the next generation of doctors who also code. This approach is not unheard of by any means, with Scale.ai and OpenAI themselves employing similar strategies over the years albeit perhaps to a lesser degree.
AI seer for breast cancer risk earns historic first ever FDA clearance
Startup Clairity secured the FDA’s first-ever clearance for an AI tool that predicts a woman’s five-year breast cancer risk from a standard mammogram. Dubbed Clairity Breast, the system analyzes subtle imaging features in mammograms—patterns even expert eyes can miss—to generate a personalized risk score for each patient. Trained on millions of images and validated on 77,000 scans across diverse centers, the model aims to flag high-risk individuals earlier than traditional risk models (which typically rely on age or family history). Importantly, it slots into the radiologist’s workflow rather than replacing it: the AI’s risk report is delivered after the radiologist’s read, guiding follow-ups like extra MRI screening for those who need it. This marks a new era of predictive screening, though the company notes the tool is an aid, not a standalone decision-maker, in breast cancer prevention.
AI platform discovers a conserved dengue virus target, paving the way for a universal vaccine
The Canadian NASDAQ-listed company, ImmunoPrecise Antibodies (IPA) claims an AI-driven breakthrough in the quest for a dengue vaccine effective against all four virus strains. Using its in-house platform LENSai – powered by a pattern-finding tech called HYFT – the company in silico sifted through viral data and identified a tiny viral epitope common to all dengue serotypes. This highly conserved snippet of the virus could become the target for a “universal” dengue vaccine, potentially sidestepping dengue’s usual trick of requiring four separate immunizations for full coverage. The discovery was done entirely via algorithms: HYFT maps biologically meaningful sub-sequences across genomes, making it possible to spot an unmutated, shared vulnerability in the virus that nature hasn’t erased through evolution. While this finding still needs experimental validation, IPA’s CEO is hailing it as “a new frontier in AI-driven biology,” illustrating how combining structural predictions with deep learning can reveal vaccine targets that human researchers might overlook. The approach might extend beyond dengue too – the team suggests similar AI tactics could tackle other pesky viruses (like HIV or norovirus) and even cancer targets where a unifying weak spot is the holy grail
Fleming Initiative & DeepMind launch AI fellowship to target antimicrobial resistance (AMR)
The Imperial College London’s Fleming Initiative has partnered with Google DeepMind to sponsor a postdoctoral fellowship focused on AI and antimicrobial resistance. This Google DeepMind Academic Fellowship will give an early-career researcher free rein to explore how AI might combat the rise of drug-resistant infections – whether by discovering new antibiotics with machine learning or deploying AI for quicker superbug diagnostics. This is historically overlooked area of medical research largely due to the drug class's unappealing economics when compared to treatments for chronic diseases. The fellow will be embedded in Imperial’s top-ranked computing department and the Fleming Initiative’s AMR research network, and even get a dedicated mentor from DeepMind’s team. Notably, Imperial is the first institution to host two DeepMind fellows at once (they already have one in computer science), a testament to the university’s emphasis on AI in science. The backdrop here is the looming AMR crisis – by 2050, drug-resistant bugs could kill millions annually if we don’t innovate. By investing in fresh talent now, DeepMind and Imperial are betting that new AI-driven approaches (from clever algorithms to predict resistance patterns to novel drug designs) could help avert that crisis. It’s a small but meaningful step to inject cutting-edge tech into one of medicine’s toughest battles.
Recursion lays off 20% of staff after axing several drug programs to refocus efforts
Recursion Pharmaceuticals is cutting around 20% of its workforce as it streamlines operations after its ranks swelled to ~800 employees after acquiring fellow AI-drug-discovery firm Exscientia last year. This comes after announcing it would deprioritize multiple pipeline programs in May and concentrate on core areas in oncology and rare diseases, effectively shelving many of the assets it acquired. Along with shelving three clinical-stage assets (for NF2, CCM, and C. diff infections) and pausing others, Recursion now says the headcount reduction will help stretch its cash runway into 2027. The layoffs, disclosed in an SEC filing, will cost about $11 million in severance. On the upside, the company still expects to end the month with ~$500M in the bank – enough to keep its slimmed-down R&D operation going for a couple more years. It’s a notable pullback for a flagship AI-driven biotech that, not long ago, touted ambitious plans (recall that post-merger, they talked of 10+ clinical readouts in 18 months). Recursion’s retrenchment suggests that even with cutting-edge algorithms, drug development remains a long, cash-hungry game where focus and efficiency are key.
Amplify Partners unveils $900M in new funds – including a $200M dedicated to techbio
Early-stage VC Amplify Partners just announced $900 million across three new funds, and notably one of them is entirely focused on tech-driven biology. The firm’s Fund 6 includes Amplify Bio, a $200M fund for “digital biology” founders with a strong focus on technical founders that – essentially, entrepreneurs as comfortable pipetting in a lab as coding in Python. It’s Amplify’s first foray into a dedicated life sciences fund, signaling how hot the computational biology and biotech software space has become. In fact, they’ve brought on a well-known computational biologist, Elliot Hershberg, as a partner to help lead this initiative. Amplify has built its reputation backing technical founders in areas like cloud infrastructure and AI, so this move suggests they see “biology x software” startups as the next big value creators. For the techbio community, it means more capital – and tech-savvy support – available at the earliest stages. Amplify Bio’s motto is literally supporting founders “in code as much as in the lab,” and with $200M to deploy, we can expect a flurry of new startups getting shots at building the AWS of biology, GitHub for cells, or other once-unsexy lab problems now ripe for software disruption.
🗓️ Upcoming BiB Events
See our handy dandy Lu.ma event calendar HERE, please RSVP so folks can plan accordingly!
Bits in Bio <> Toronto Tech Week Happy Hour, June 24th at 6:30 pm
📰 Top Community Conversations
Its a tough job market out there, what bas pay would you guess for this role?
🏢 New Job Openings
In Vivo Scientist/Senior Scientist at Epi Labs
Postdoctoral Researcher at EPFL
Various Roles at Transfyr
Senior Data Scientist at Pioneer Labs
🙋 🙏 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!