Bits In Bio Weekly - April 15th 2025
Updates From and Around the Bits in Bio Community
📢 Highlights
FDA Embraces AI, Human Organoids, and Digital Twins Over Traditional Animal Testing
Baker Lab's New RFdiffusion2 Nails All 41 Test Cases in Enzyme Design Benchmark
GenomOncology's Open-Source BioMCP Gives LLMs Real-Time Biomedical Data Access
Not yet a member of our super awesome slack community of >8500? Join HERE 🤗
👀 In Case You Missed it …
FDA Bids Farewell to Lab Animals, Embraces AI Models and Organoids
In a landmark move, the FDA is modernizing drug development by phasing out animal testing requirements for monoclonal antibodies and other new drugs. In its place comes a toolkit of high-tech alternatives: AI-driven simulations and human-derived models (like organoids and organs-on-chips) that mimic human biology more directly than mice ever could. The FDA's animal testing phaseout initiative represents one of the first major policy shifts under Commissioner Martin Makary's leadership, with pharmaceutical companies soon leveraging computational models and organ-on-chip technologies to simulate how monoclonal antibodies distribute through the human body and predict potential toxicity issues. This regulatory evolution comes after the passage of the FDA Modernization Act 2.0 in 2022, which for the first time authorized using non-animal alternatives in preclinical testing but had seen limited industry adoption due to regulatory compliance concerns. The plan promises safer, faster drug development by using advanced computational toxicity predictions and lab-grown human tissues instead of live animal trials. Starting with new IND applications, the agency will encourage data from these cutting-edge methods and even lean on real-world human data from abroad to avoid redundant animal studies. Companies that provide strong non-animal safety evidence may get quicker reviews, hinting at cheaper R&D and cures arriving sooner. It’s a paradigm shift in regulatory science that could make drug testing more humane and efficient worldwide.
Baker Lab's New AI Creates Custom Enzymes from Atomic Blueprints
The Institute for Protein Design at University of Washington has released RFdiffusion2, a breakthrough AI system that creates enzymes from minimal atomic specifications without requiring pre-defined rotamers or index positions, solving key limitations in existing enzyme design methods. The system builds on collaborative work from the Baker Lab and DiMaio Lab at IPD along with the Barzilay Lab and Jaakkola Lab at MIT, using deep learning to design protein structures precisely tuned for specific chemical reactions with unprecedented accuracy. In in-silico testing, RFdiffusion2 solved all 41 cases in a challenging enzyme design benchmark (the Atomic Motif Enzyme benchmark based on the M-CSA database), dramatically outperforming previous tools that solved only 16 cases. Laboratory validation confirmed the model's effectiveness by successfully producing active enzymes for five distinct chemical reactions with fewer than 100 designs tested per case—a stark departure from traditional workflows requiring screening of thousands of candidates—including retroaldolase, cysteine hydrolase, and zinc hydrolase designs, with the latter exhibiting orders-of-magnitude higher activity than previously engineered metallohydrolases. The innovation enables a broader range of active site geometries and unlocks new applications in enzyme design, potentially accelerating development of custom enzymes for applications like plastic degradation, drug manufacturing, and other complex chemical reactions.
BioMCP Empowers LLMs with Standardized Access to Biomedical Data
Ohio-based GenomOncology has announced their open-source project called BioMCP, which tackles a major AI blind spot by giving research assistants like GPT-style models real-time access to biomedical data. Launched by GenomOncology’s Ian Maurer, the BioMCP server hooks into public NIH APIs (ClinicalTrials.gov, PubMed/PubTator, MyVariant.info, etc.) so an AI can pull in up-to-date clinical trial info, papers, and genomic variant data as context. The idea is to prevent the “stale knowledge” problem — ensuring an AI helper in precision oncology always consults the latest evidence instead of just its fixed training data. A demo shows an AI querying trial criteria and genetic databases on the fly via BioMCP, hinting at more interactive, data-aware medical chatbots. Released under MIT open source license, you can find it here.
Startup Launches with $121M to Wipe Out Disease-Causing Antibodies
Newcomer Merida Biosciences has emerged with a $121 million Series A to develop therapies for autoimmune and allergic diseases by literally removing the problem-causing antibodies. Merida’s platform creates engineered proteins (built on an IgG Fc scaffold) that latch onto only the bad antibodies – like those that drive Graves’ disease or severe allergies – and tag them for destruction by the body, while leaving normal immune function alone. The approach promises to be more precise and durable than today’s blunt immunosuppressants: imagine treating an autoimmune disease by knocking out just the harmful antibody (and the B-cells that produce it) instead of nuking the whole immune system. Their lead candidate targets Graves’ disease by neutralizing the rogue antibodies attacking the thyroid, potentially restoring healthy metabolism without the usual drastic measures. With backing from Third Rock, Bain, and others, Merida is betting that a cleaner fix to immune malfunctions is within reach – one that could tackle dozens of antibody-driven conditions that have long evaded selective treatment.
PTM-Mamba: Enhancing Protein Language Modeling with Post-Translational Modification Integration
Researchers have unveiled PTM-Mamba is a newly developed protein language model developed by at Duke University that addresses a critical gap in current protein modeling by explicitly representing post-translational modifications (PTMs), which vastly expand protein functional diversity and are often implicated in diseases. The model integrates PTM tokens using bidirectional Mamba blocks—a modern neural network architecture that efficiently processes sequential data in both forward and backward directions—fused with ESM-2 protein language model embeddings via a specially designed gating mechanism that maintains comprehension of both wild-type and modified protein sequences. The team utilized multiple protein language models including SaLT&PepPr, PepPrCLIP, and PepMLM to generate the targeting peptides. PTM-Mamba outperforms baseline models on several key tasks including disease association prediction, druggability assessment, and determining how PTMs affect protein-protein interactions. The model also shows promising capability for zero-shot PTM discovery, where it can predict plausible modifications for specific residues without additional training. This research has potentially significant applications in understanding disease mechanisms and improving therapeutic design with enhanced targeting specificity. The model is available on HuggingFace and code can be found on Github.
IngeniX.AI Secures $9M to Simulate Clinical Trials
Poland-based IngeniX.AI has raised $9 million to advance its multimodal AI foundation model for simulating clinical trials, potentially transforming how pharmaceutical companies evaluate drug candidates before committing to expensive human studies. The platform integrates diverse patient data—including genomics, proteomics, clinical records, and real-world evidence—to create virtual cohorts and simulate treatment responses across different populations, enabling researchers to rapidly iterate trial designs while identifying optimal endpoints and inclusion criteria. Founded by veterans from both AI development and pharmaceutical R&D, the company claims its simulations can predict clinical outcomes with up to 85% accuracy for certain therapeutic areas, potentially saving sponsors millions in development costs while reducing the ethical concerns associated with exposing patients to experimental therapies that might fail. The funding round was led by a mix of biotech-focused venture capital firms and strategic pharmaceutical partners, with plans to expand the platform's capabilities across additional therapeutic areas while pursuing validation studies comparing simulation predictions to actual trial outcomes. IngeniX.AI joins a growing field of computational trial simulation providers, though it distinguishes itself through its foundation model approach that continuously learns from new data inputs while optimizing for specific disease biology and drug mechanisms.
Recursion's AI and Enamine Curates 15K Molecules for 100 Tough Drug Targets
Salt Lake City Recursion Pharmaceuticals and Enamine have unveiled AI-curated screening libraries targeting 100 historically challenging drug targets, leveraging machine learning to identify novel chemical matter for proteins previously deemed "undruggable." The collaboration combines Recursion's ML-powered prediction platform with Enamine's billion-compound REAL (Readily Accessible) database to generate focused libraries of 3,000-5,000 compounds for each target, dramatically reducing screening inefficiencies compared to traditional high-throughput approaches. Among the targets are protein-protein interactions, transcription factors, RNA-binding proteins, and other challenging mechanisms that have resisted conventional drug discovery methods despite their therapeutic potential across multiple disease areas. Recursion's predictive models were reportedly trained on the company's proprietary biological data, including cellular morphology changes detected through high-content imaging across tens of millions of experiments. The libraries will be available to pharmaceutical partners seeking to jumpstart development programs against these difficult targets, potentially expanding the universe of druggable biology while accelerating the early discovery phase for multiple therapeutic areas.
ImmunityBio Gets $75M Lifeline from New Investor
A leading immunotherapy biotech, ImmunityBio, has secured a $75 million equity infusion from a single institutional investor – a welcome lifeline for its pipeline of cancer and vaccine programs. The deal involves a direct stock sale (with warrants that could bring in an extra $90M later) to shore up working capital at a time when the company was rumored to be cash-strapped. It comes on the heels of ImmunityBio’s first product approval – an IL-15 based therapy called Anktiva for bladder cancer – and will help fund ongoing trials and manufacturing. Large injections of capital like this are often a sanity check for struggling biotechs, suggesting that at least one big backer sees promise in ImmunityBio’s approach to activating the immune system against disease. With the new funds, the company can keep pushing forward on its natural killer cell and T cell-based treatments instead of worrying about its runway.
23andMe’s Genetic Trove Pegged at $289M Amid Bankruptcy Proceedings
A new analysis from data valuation firm Gulp Data puts a $289 million price tag on 23andMe’s genetic database just as the direct-to-consumer DNA company grapples with potential bankruptcy. 23andMe has millions of customers’ DNA data paired with health information – a goldmine for pharmaceutical research if it were ever sold – though privacy concerns loom large. Gulp Data arrived at the figure using market comps and proprietary algorithms, underscoring how valuable big biomedical datasets can be even when a company’s finances falter. The situation spotlights a thorny question: if 23andMe goes under, could its trove be treated like an asset to be bought, and what would that mean for customers who consented to give their DNA? At the very least, this valuation is a reminder that in biotech, data itself is often the crown jewel – something investors (and regulators) are watching as much as any drug or device.
Recursion's AI-Designed Molecule Reaches Human Testing In Record Time
Recursion Pharmaceuticals has dosed the first patient in a Phase 1 trial of REC-2554, an AI-designed MALT1 (Mucosa-Associated Lymphoid Tissue Lymphoma Translocation Protein 1) inhibitor targeting both B-cell malignancies and autoimmune disorders. The small molecule, discovered using the company's Recursion OS platform and massive biological dataset, represents a significant milestone for AI-driven drug discovery by successfully translating computational predictions into clinical-stage development for a valuable oncology target. MALT1 functions as both a scaffold protein and a protease in the NF-κB signaling pathway, making it a compelling but complex target for intervention in conditions with immune dysregulation, with preclinical data showing REC-2554 has potent anti-tumor activity and favorable pharmacokinetic properties. Now in a Phase 1 study (dubbed EXCELERIZE), researchers will evaluate its safety, optimal dose, and initial efficacy both alone and later in combination with other treatments. It’s a milestone for AI-driven drug discovery: a computationally-designed cancer drug advancing in the clinic, potentially faster and more precisely tailored than traditional chemistry could manage.The open-label Phase 1 trial will evaluate the safety, tolerability, pharmacokinetics, and preliminary efficacy of escalating doses in patients with relapsed or refractory B-cell malignancies, with plans to expand into autoimmune indications following initial safety validation.
🗓️ Upcoming BiB Events
See our handy dandy Lu.ma event calendar HERE, please RSVP so folks can plan accordingly!
.Bits in Bio April Journal Club - Drug Repositioning, Wednesday, April 23 at 5:30 PM PT
Thank you to Steve Cheney for organizing!
Chicago Bits in Bio Happy Hour, Tuesday, May 29 at 6:00 PM
First event in Chicago!
Thank you to Matthew Schechter for organizing
📰 Top Community Conversations
We've launched our new super slick website at www.bitsinbio.org!
A big thanks to the team at www.wunderdogs.co for their excellent work.
🏢 New Job Openings
Notable postings below - over 100 more on our community job board!
Full-Stack Engineer at ModernVivo
(Founding) Computational Biologist at Differential Bio
(Founding) Full-Stack Software Engineer at Differential Bio
🙋 🙏 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!