📢 Highlights
China's XtalPi and DoveTree ink largest AI-pharma deal yet at $5.99 billion
RosettaFold-3 beats AlphaFold on protein chirality while staying fully open-source with the modular AtomWorks AI toolkit
From rare to routine: Kriya banks $313M to democratize gene therapy manufacturing for millions, not just thousands, of patients
Not yet a member of our super awesome slack community of ~10,000? Join HERE 🤗
✨ Special Announcement! ✨
I'm excited to share with you that we have launched the Bits In Bio Podcast Series!
Join your hosts Robbie Matthews and Vincent Alessi every month for deep dive into Science x AI with the builders-turned-executives leading the charge.
We are kicking off with John Axerio-Cilies, the Chief Data Officer of Tempus AI (NASDAQ: TEM), talking about his journey and whats to come in the rapidly moving field of data at the intersection of care delivery and therapeutic development.
Find it on Apple Podcasts and on Spotify
Join us next episode with Apheris CEO Robin Röhm, who's built the federated learning infrastructure that's letting the landmark OpenFold Consortium tap into billions of data points from big pharma's jealously-guarded structural information.
👀 In Case You Missed it …
Chinese AI platform XtalPi lands record $6 billion DoveTree partnership
XtalPi and DoveTree Medicines, founded by Harvard's Gregory Verdine, announced a landmark $5.99 billion strategic collaboration for AI-driven drug discovery—potentially the largest AI-pharma deal to date and a watershed moment for computational drug development. The deal structure includes $51 million upfront to XtalPi, an additional $49 million within 180 days, and up to $5.89 billion in milestones and royalties, with XtalPi's platform integrating quantum physics predictions, AI-driven molecular design, and robotics to tackle "undruggable" targets including RAS, Myc, and β-catenin across oncology, immunology, inflammatory, neurological, and metabolic indications. XtalPi, founded in 2015 by MIT-trained scientists and previously raising $784 million including a $400 million Series D led by OrbiMed, saw its stock surge 12.4% on the announcement as markets recognized the validation from Verdine—who co-founded 12+ biotechs including Enanta Pharmaceuticals and co-developed three FDA-approved drugs. The collaboration employs a multimodal approach spanning small molecules, biologics, antibody-drug conjugates, and molecular glues, with Verdine noting XtalPi's platform "has the potential to transform the profound uncertainties of drug discovery into quantifiable engineering solutions." Following biotech partnership models but at unprecedented scale, the deal highlights exceptional confidence in AI platforms' ability to generate blockbuster drugs without requiring their own clinical infrastructure—a bet that computational approaches can finally crack biology's most intractable targets. From a macro lens, this can be seen as a clarion call for the growing trend of biotech AI excellence being relocated outside of the US.
AtomWorks framework and RosettaFold-3 to democratize biomolecular structure prediction with open-source tools
Researchers from the Institute for Protein Design at the University of Washington and collaborators have announced AtomWorks, a modular computational framework that standardizes data processing for biomolecular machine learning models, solving the heterogeneous data quality and annotation challenges that have fragmented open-source structural biology efforts. The framework's modular architecture allows researchers to reuse over 80% of code across different applications—from structure prediction to protein design—dramatically reducing the need to rebuild data pipelines for each new model. Using AtomWorks, the team trained RosettaFold-3 (RF3), which incorporates improved chirality representations through geometric features and atom-level conditioning capabilities, achieving 88% accuracy for ligand chiral centers compared to 84% for AlphaFold3 and 76% for Boltz-2. When evaluated on recent PDB structures, antibody-antigen complexes, and mixed L/D peptides, RF3 consistently outperformed other open-source alternatives like Boltz while approaching AlphaFold3's accuracy across protein-protein, protein-ligand, and protein-nucleic acid interactions. The framework successfully handles the full complexity of biomolecular data, correctly processing covalent modifications, multiple occupancies, and structures with symmetry centers—edge cases that often require specialized expertise in biochemistry and structural biology. RF3's atom-level conditioning enables users to specify desired distances between atoms for incorporating experimental constraints or performing protein docking tasks. The complete AtomWorks framework, RF3 model weights, and curated training datasets are available on GitHub at https://github.com/RosettaCommons/atomworks and https://github.com/RosettaCommons/modelforge under a permissive BSD license.
Gene therapy biotech Kriya banks $313M for common disease gene therapies
Durham and Palo Alto-based Kriya Therapeutics raised $313 million in new funding per SEC filings, bringing total funding to over $600 million since its 2019 founding by CEO Shankar Ramaswamy (brother of Roivant's Vivek Ramaswamy) alongside Spark Therapeutics co-founder Fraser Wright and ex-United Therapeutics executive Roger Jeffs with founding investor Transhuman Capital led by CR Sincock. Kriya's contrarian strategy targets gene therapies for common diseases affecting millions—not just rare diseases—with focus areas spanning ophthalmology (lead program KRIYA-586 for thyroid eye disease), metabolic disease, and neurology, supported by in-house manufacturing capabilities designed to address the industry's persistent capacity bottleneck. The funding extends Kriya's cash runway into late 2026 and follows strategic acquisitions of Redpin Therapeutics (epilepsy, trigeminal neuralgia programs) and Tramontane Therapeutics (NASH program), building what Ramaswamy calls a "consolidated engine" approach distinct from his brother's traditional biotech model. With multiple programs advancing toward clinical trials and leadership expertise from successful gene therapy launches including Luxturna and Zolgensma, Kriya aims to democratize gene therapy access through manufacturing scale—a critical differentiator as the field shifts from ultra-rare diseases toward "highly prevalent diseases affecting millions." The timing is opportune as gene therapy moves past early setbacks toward broader applications, though the $313 million raise in this market suggests either exceptional investor confidence or concerning cash burn rates.
Fast-acting psychedelic from Reunion meets Phase 2 endpoint in postpartum depression
New Jersey-based Reunion Neuroscience's psychedelic prodrug RE104 met its primary endpoint in the Phase 2 RECONNECT trial for postpartum depression, with the 30mg dose achieving a 23.0-point MADRS reduction versus 17.2 points for the 1.5mg active control (p=0.0094) in 84 women with moderate-to-severe PPD across 38 U.S. clinical sites. RE104, a prodrug of 4-OH-DiPT offering a 3-4 hour psychedelic experience versus psilocybin's longer duration, demonstrated 77.1% response rate at Day 7 with effects sustained through Day 28, while maintaining favorable tolerability with nausea (43.9%) and headache (34.1%) as primary adverse events. The success is particularly notable given PPD affects approximately 15% of new mothers (~500,000 diagnosed annually in the U.S.) with only two approved treatments—Sage's struggling Zurzuvae and the discontinued Zulresso—leaving massive unmet need for rapid-acting interventions. Owned by MPM BioImpact following a ~$13 million acquisition in 2023 and backed by $103 million from MPM Capital and Novo Holdings, Reunion plans Phase 3 trials for 2026 plus an additional Phase 2 in adjustment disorder, positioning RE104's shorter duration as more practical for clinical implementation than traditional psychedelics. Lead investigator Dr. Anita Clayton from University of Virginia notes that "current treatment options for PPD remain limited and are often slow-acting and burdensome"—understating the crisis where new mothers often wait weeks for antidepressants to work while battling thoughts of self-harm.
Insilico Medicine's AI platform speeds NLRP3 inhibitor toward Parkinson's trials
Biotech-AI iconoclast Insilico Medicine completed IND-enabling studies for ISM8969, an oral NLRP3 inflammasome inhibitor targeting Parkinson's disease, with plans to submit IND application in Q4 2025 following favorable preclinical results showing dose-dependent motor function improvements approaching healthy control performance in animal models. The blood-brain barrier penetrable small molecule, generated using Insilico's proprietary Pharma.AI platform that achieves 12-18 month development candidate timelines versus traditional 2.5-4 years, demonstrated significant improvements across three behavioral tests (open field, rotarod, grip strength) at the optimal 20 mpk dose. This represents a potential paradigm shift from symptomatic to disease-modifying PD treatment by targeting underlying inflammatory progression rather than just dopamine replacement—addressing a disease affecting 25.2 million globally by 2050 with current treatments offering mainly symptomatic relief with limiting adverse effects. ISM8969 joins Insilico's portfolio of 30+ AI-designed assets with 22 development candidates nominated since 2021 and 10 IND clearances, with CSO Feng Ren noting the progress "strengthens our belief in the potential of ISM8969, as well as the reliability of AI-driven drug discovery method, in the new field of CNS diseases." CEO Alex Zhavoronkov emphasizes that "the anti-inflammation strategy against Parkinson's disease has never before passed the whole process of clinical validation"—making ISM8969's advancement particularly encouraging for both the platform and the therapeutic approach.
Clinical Trial shows Deep learning model with Breast Cancer Imaging reduced unnecessary biopsy
In an international multicenter clinical trial, a deep learning model using shear wave elastography (AI-SWE) was tested on breast masses classified as “suspicious” or “probably benign” on ultrasound. Shear wave elastography (SWE) is an advanced ultrasound technique that measures tissue stiffness and has been investigated as a complement to breast ultrasound. Trained on over 4,000 images from 11 institutions, the system matched expert accuracy in detecting malignancy and reduced false positives from 53% to 39%, suggesting it could help lower unnecessary biopsies. Its sensitivity was 97.8%, comparable to that of human experts.
ProteomeLM AI predicts protein–protein interactions (and essential genes) at scale
A new preprint from EPFL Switzerland unveils ProteomeLM, a massive language model trained on protein sequences that can predict which proteins bind up and which genes are truly essential. The model essentially acts like a matchmaking service for the cell: given a whole proteome, it ranks all possible protein–protein interactions in one go, rather than testing pairs one by one. In early tests, ProteomeLM’s “single pass” predictions could retrieve ~70% of known E. coli protein interactions by just looking at the top 5% of its rankings (a huge boost in efficiency). They fine‑tune this base model into two heads: ProteomeLM‑PPI, which predicts protein–protein interactions faster than coevolutionary methods while generalizing to novel species, and ProteomeLM‑Ess, which flags essential genes with cross‑taxa accuracy. Intriguingly, the AI can also flag likely essential genes across different species, hinting at a computational shortcut to pinning down which genes an organism can’t live without. The upshot? If these results hold up, labs may soon have an AI assistant to help map cellular interaction networks and critical genes – accelerating everything from drug target discovery to understanding the fundamental “wiring diagram” of life.
Morphological map of under- and overexpression of genes in human cells
The JUMP Cell Painting Consortium has delivered one of the most comprehensive morphological map of human gene function to date, perturbing roughly 75% of the protein-coding genome in human U-2 OS cells and generating single-cell images across 15,243 genes. The dataset includes overexpression profiles for 12,609 genes using open reading frames and knockout profiles for 7,975 genes via CRISPR-Cas9, creating a rich resource of phenotypic data that captures previously undiscovered gene clusters and functional relationships. This work addresses a bottleneck in biological research: while we can sequence genomes at scale, connecting genetic variation to cellular phenotypes has remained largely labor-intensive. By providing standardized morphological readouts for a great portion of human protein-coding genes, this dataset can enable researchers to rapidly generate hypotheses about gene function, identify potential drug targets, and understand how genetic perturbations manifest at the cellular level. All data has been released for the scientific community via the Cell Painting Gallery on AWS, with well-level profiles, image analysis pipelines, and over 300,000 single-cell measurements made publicly available
SkyHawk snaps a $2B deal with Merck KGaA, splicing modulators are in
Merck KGaA (the German Merck) is paying up to $2 billion in a new partnership with Skyhawk Therapeutics to go after neurological diseases using Skyhawk’s RNA-splicing modulation tech. The company licensed their Hungtington small molecule splicing modulator to Merck for up to $2B in milestones. This is not their first try, with previous deals with Ipsen, Vertex, Genentech, Takeda, CellGene, and Biogen. This is not the first company making big deals with splicing modulators in the CNS, as PTC also got a billion dollar deal with Novartis for the same target last year. While splicing modulation is generally targeted with ASOs (e.g. Spinraza for SMA), PTC paved the way with a small molecule that alters splicing factors in a rather targeted way (risdiplam also for SMA). Skyhawk has a who’s-who list of existing partners – Merck & Co. (yes, the other Merck), Vertex, Takeda, Ipsen – so this new alliance adds Germany’s Merck to the RNA modulator fan club
Lilly inks $1.3B partnership with Superluminal Medicines to hunt down new obesity drugs
Indianapolis-based Eli Lilly signed a $1.3 billion collaboration with Boston's Superluminal Medicines to discover AI-driven small-molecule therapeutics targeting G protein-coupled receptors (GPCRs) for cardiometabolic diseases and obesity—a strategic move following disappointing Phase III results for its oral GLP-1 orforglipron (12.4% weight loss versus competitors' 13.7%). The deal grants Lilly exclusive rights to develop and commercialize compounds from Superluminal's proprietary AI platform that integrates machine learning with protein dynamics and structural biology to unlock the 70% of GPCRs that remain "undrugged" despite the receptor class comprising 35% of all approved medicines. Superluminal, founded in 2022 and backed by $120 million from RA Capital, Insight Partners, and NVIDIA's NVentures, brings a lead MC4R agonist program expected to enter clinic in 2026 (retained outside the Lilly deal) along with four additional GPCR programs in lead generation. The partnership structure—upfront payments, equity investment, development milestones, and tiered royalties—follows Lilly's recent spending spree including the $856 million Gate Biosciences acquisition and reflects the pharma giant's determination to maintain dominance in the $150 billion obesity market projected by decade's end. It's another example of big pharma placing massive bets on AI platforms to accelerate discovery timelines, with Superluminal's CEO Cony D'Cruz noting that while GPCRs have proven themselves in obesity and cardiometabolic landscapes, "we're at the very early stages of exploration of the target class
Chai Discovery's antibody AI earns $70 million Series A
San Francisco-based Chai Discovery announced a $70 million Series A led by Menlo Ventures (including their Anthropic-partnered Anthology Fund) to advance AI-driven molecular design following breakthrough achievements with their Chai-2 model achieving near-20% hit rate in fully de novo antibody design—a 200x improvement over the previous 0.1% computational state-of-art. The round, bringing total funding to $100 million, includes Yosemite, DST Global Partners, SV Angel, Avenir, DCVC, plus existing investors Thrive Capital, OpenAI, Dimension, and adds former Pfizer CSO Mikael Dolsten to the board—who advanced 150 molecules to trials with 36 approved medicines during his tenure. Chai-2 requires only target antigen and epitope information to generate successful binders, solving in two weeks what one partner spent three years and $5+ million attempting, fundamentally transforming drug discovery from "searching a giant bunch of keys" to having "a master locksmith design exactly the right shape key," as co-founder Matthew McPartlon describes. Founded in 2024 by ex-Absci/Facebook AI/OpenAI researcher Joshua Meier alongside Stripe engineering leader Jack Dent, the company released the open-source Chai-1 foundation model before unveiling Chai-2 in July 2025, with Menlo's Greg Yap noting "a meaningful fraction of the biotech industry already applied for Chai-2 access." The technology enables targeting previously inaccessible therapeutic targets while dramatically accelerating development timelines—representing a shift from trial-and-error experimentation to designed engineering that Meier believes will finally make biology tractable.
Convoke Bio gets out of stealth with a $8.6m raise from Kleiner Perkins and Dimension
While a lot of the action is currently focusing on leveraging AI for drug design, or clinical trial optimization, the rest of the work in pharma has been relegated to the sidelines.
Convoke aims to address that issue with its $8.6m raise, and help speed up the move from idea to drug, by streamlining regulatory submissions, investigators recruitment, clinical trial submissions, target product profiles, etc…Convoke promises they will build an "AI-native operating system for biopharma" to automate the knowledge work that consumes 45% of drug development timelines—where each day of delay costs $500,000 in lost sales. The platform unifies internal and external data into a searchable ontology while deploying custom agents for regulatory documents, competitive intelligence, and manuscript generation—essentially providing "digital colleagues" for literature review and report creation. The investor syndicate includes ACME, Comma Capital, Liquid2, Not Boring Capital, Audacious, and Lux Capital, plus angels Qasar Younis and Erik Torenberg, betting that Convoke can democratize pharma expertise for smaller organizations. Early customers range from small biotechs to top-20 pharma companies, with the team comparing their potential impact to how cloud computing transformed software development—a particularly apt analogy given that Paul Janssen founded a pharma company with ~$12,000 in the 1950s while today's startups burn millions just on planning. The timing is strategic as biopharma drowns in exponentially growing data while facing pressure to accelerate timelines and reduce costs. Part of the team is Alex Telford whose blog is a recommended read to anyone, the company was started following his deep dive on what made Janssen a successful pharma, and bringing what he believed made it great into a product available to all pharma.
VantAI scores billion-dollar RIPTAC deal with Yale spinout Halda
New York-based VantAI and New Haven's Halda Therapeutics announced a strategic alliance worth over $1 billion to discover next-generation RIPTAC (Regulated Induced Proximity Targeting Chimeras) medicines, leveraging VantAI's Neo-1 foundation model and NeoLink structural proteomics platform to identify optimal target-effector pairs for oncology and immunology. Founded by Yale's Craig Crews—the PROTAC pioneer who literally invented targeted protein degradation—Halda's RIPTAC approach uses bifunctional molecules that bind both tumor-specific proteins and essential cellular proteins, creating a "hold and kill" mechanism distinct from traditional degraders that simply eliminate proteins. The deal structure includes upfront payments, research support, development/commercial milestones, and tiered royalties, with Halda's lead candidate HLD-0915 already in Phase 1/2 trials for metastatic castration-resistant prostate cancer and sporting FDA Fast Track designation (NCT06800313). VantAI's AI platform addresses the combinatorial challenge of matching thousands of potential target-effector pairs—a problem intractable for traditional screening but perfectly suited for machine learning approaches that can predict protein-protein interactions at scale. This represents the "second wave" of proximity-based therapeutics following PROTACs' success, with CEO Christian Schade noting the partnership will unlock "previously inaccessible targets" through AI-driven molecular design.
Hatteras defies VC slowdown with $200M+ for early-stage life sciences
Hatteras Venture Partners closed more than $200 million in capital commitments across two healthcare-focused funds—Hatteras Venture Partners VII and the new Hatteras Opportunity Fund I—bringing total assets under management to $900+ million across seven funds over the firm's 25-year history. The fundraising success is notable amid industry-wide slowdown where median fund closure time has stretched to 15.3 months (longest in a decade) and H1 2025 saw only $26.6 billion raised across 238 funds, a near decade-low pace for venture capital deployment. Hatteras has backed 100 portfolio companies since inception including notable exits G1 Therapeutics, HistoSonics, Kymera Therapeutics, StrideBio, and Ribometrix, maintaining its academic partnership model particularly with southeastern U.S. institutions where venture capital remains scarce relative to traditional biotech hubs. The firm's focus on seed- and early-stage investments across biotechnology, medical devices, and health technology provides crucial funding during challenging market conditions when many VCs have retreated to later-stage, de-risked opportunities. General Partner Clay Thorp reflected that "what has been most gratifying in this journey is the relationships we have made with scholarly innovators, committed entrepreneurs, and resilient co-investors"—a sentiment that sounds quaint but represents real differentiation in a market where coastal VCs increasingly ignore everything between Boston and San Diego.
OrbiMed gathers $1.86B for nondilutive royalty deals as biotechs flee equity markets
New York-based OrbiMed closed $1.86 billion in commitments for its fifth royalty and credit opportunities fund, focused on non-dilutive financing for growth-oriented healthcare companies desperately avoiding dilutive equity raises in today's punishing biotech market. The fund, with over 90% of capital from existing relationships, targets biopharmaceuticals, medical devices, diagnostics, and technology-enabled healthcare services through royalty streams, credit facilities, and structured solutions—essentially becoming the biotech industry's alternative to predatory equity rounds. OrbiMed, managing over $17 billion across 145 professionals globally with offices from San Francisco to Shanghai, has 134 biopharma companies in its portfolio and recently deployed capital through deals including a $54.5 million seed for Renasant Bio and $75 million private placement for Verastem Oncology. General Partner Matthew Rizzo notes their "flexible, non-dilutive capital solutions can help portfolio companies grow and thrive despite volatile market conditions"—diplomatic phrasing for what's essentially emergency financing when public markets won't touch you and VCs demand 50% of your company. The successful fundraising despite "challenging equity markets" positions OrbiMed to capitalize on the growing disconnect between private valuations and public market reality, where companies increasingly trade future revenue for present survival.
🗓️ Upcoming BiB Events
See our handy dandy Lu.ma event calendar HERE, please RSVP so folks can plan accordingly!
AI Transformation Across the Biopharma R&D Lifecycle, Wednesday August 27th in San Francisco
Thank you to Shantenu Agarwal, Eddie Eltoukhy, Emma Krivoshein, and Ashton Teng for organizing!
📰 Top Community Conversations
Not much to report this week - have a question?
Post it in our #ask-bib slack channel
🏢 New Job Openings
Senior Computational Biologist, Science at Chan Zuckerberg Initiative
Bioinformatician Data Analyst at University of Virginia
🙋 🙏 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!
Vince Alessi
BiB Editor in Chief
Subscribe for free!
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.