Artificial intelligence & biotech · RNA foundation models
Blank Bio
A San Francisco lab training RNA foundation models to help pharma companies pick the right trial patients and predict how a tumor will behave. Founded in 2025, out of Y Combinator's Summer 2025 batch, with a $7.2 million seed round and a sequencing partnership with PacBio announced in May 2026.
A model trained on four GPUs
In their Y Combinator launch post, Blank Bio's three co-founders described a model that beat rivals trained on far more hardware. Co-founder and CTO Ruian Shi wrote that he had "used 4 GPUs to pre-train an mRNA foundation model that beat competitors using months of compute on thousands of H100s." The company says that model, at 10 million parameters, is lightweight enough to score hundreds of RNA sequences in seconds, and that it already outperforms 59 other models from labs and companies working on the same problem.
The pitch is not a new drug. It is a way to read biology that already exists. Blank Bio's website says it "enables pharma teams to identify the right patients for their clinical trials and predict how their disease will progress." No product has been submitted to the FDA, and there is no clinical trial. What there is, so far, is a foundation model, a seed round, and a sequencing deal.
The problem: too much RNA, too little use of it
Blank Bio's website argues that bulk RNA sequencing has become cheap and common enough to run in clinical settings, but that most of what it captures gets thrown away. "We're starting to generate more bulk RNA-seq data than we know how to fully interpret," the company writes. "Every sample contains a rich snapshot of disease biology, but much of that information is still unused in practice." Standard pipelines, the company says, compress a sample down to per-gene counts, discarding isoform structure and other patient-specific detail.
The company's launch post frames the design problem in stark terms: for a typical mRNA therapeutic there are roughly 10 to the power of 2709 possible sequences, "unimaginably larger than the number of stars that exist in the known universe." Blank Bio's bet is that a model trained to represent RNA sequences well can substitute for some of that experimentation, in designing molecules and in reading what a patient's tumor is doing.
Three founders, one prior collaboration
Jonathan Hsu, the CEO, was an early employee at Valence Discovery, an AI for drug discovery company that spun out of the Mila research institute. He led product and strategy there. Valence was acquired by Recursion Pharmaceuticals, a Nasdaq-listed company, in May 2023, and Hsu stayed on to help build what became Recursion's AI research arm, Valence Labs. He describes himself as a biochemist by training and a graduate of the University of Waterloo.
Philip Fradkin spent four years at Deep Genomics, one of the first employees there, and holds a PhD in computational biology from the University of Toronto. Ruian Shi, the CTO, has a PhD in computer science from the same university, worked as a research scientist at Memorial Sloan Kettering Cancer Center, and was previously a software engineer at Amazon. The company's launch post says, "Phil and Jonny previously worked together at Recursion. Ian and Phil have been collaborating for 5 years throughout their PhDs, doing ML research specifically for RNA biology." Ian is the name the founders use for Ruian Shi there.
That five-year collaboration between Fradkin and Shi produced Orthrus, an RNA foundation model the two co-authored along with researchers including Brendan Frey, Quaid Morris, Leo Lee and Bo Wang. The preprint was posted to bioRxiv in October 2024, before Blank Bio existed as a company, and was later published in Nature Methods. Blank Bio's own funding announcement calls Orthrus "its prior academic work on RNA foundation models."
From launch to a seed round and a sequencing deal
Blank Bio launched publicly on Y Combinator in mid 2025, in the Summer 2025 batch, with six people and a working model. The launch post said the open-source model was "already being used by Sanofi and GSK," and that Blank Bio was collaborating with the Arc Institute on virtual cell models. Neither claim has independent confirmation.
On May 19, 2026, Blank Bio announced the close of a $7.2 million seed round, describing it as oversubscribed, with Define Ventures, Leonis Capital, Nova Threshold, Ripple Ventures, SignalFire, Y Combinator, "and others." No lead investor was named. The same announcement disclosed a strategic collaboration with Pacific Biosciences: Blank Bio will generate PacBio HiFi long-read RNA sequencing data from up to 100 fresh frozen tumor samples, spanning multiple cancer types, sequenced at Seattle Children's Research Institute on an automated Kinnex library platform. A separate SEC filing shows an Alumni Ventures vehicle organized around Blank Bio had already raised money from eight investors by November 2025, about six months before the announcement.
What is proven, and what is still claimed
| Evidence | What the record shows | Source type |
|---|---|---|
| Orthrus foundation model | Preprint posted to bioRxiv Oct 10, 2024, co-authored by Fradkin and Shi with five other researchers. Indexed on PubMed (PMID 39416135) and later published in Nature Methods. Describes a Mamba based RNA foundation model trained with a contrastive learning objective, evaluated on five mRNA property prediction tasks. | Public record |
| Model performance claims | Company says its current model outperforms 59 other RNA foundation models and runs predictions on hundreds of sequences in seconds. No independent benchmark paper naming Blank Bio's production model was found as of Sep 25, 2026. | Company stated |
| Sanofi and GSK usage | Company says its open source model "is already being used by Sanofi and GSK." No statement from either pharma company, or independent report, confirms this. | Not found |
| Arc Institute collaboration | Company says it is "collaborating with the Arc Institute to bring our RNA embeddings into the next generation of virtual cell models." No confirmation from the Arc Institute was found. | Not found |
| PacBio collaboration | Announced jointly with PacBio on May 19, 2026. Up to 100 tumor samples, sequencing at Seattle Children's Research Institute. Quoted by PacBio's own VP of marketing in the same release. | Company and partner stated |
| Seed financing | $7.2 million, described as oversubscribed and closed by May 19, 2026. No Form D found under Blank Bio's own name in SEC EDGAR as of Sep 25, 2026, only a $135,000 SPV formed to invest in the company, filed six months earlier. | Differs from filing record |
| FDA, clinical trials, NIH | None found. The company has not submitted a device or diagnostic for clearance and is not running a registered clinical trial. | Not found |
Read plainly: the science behind Blank Bio has a public paper trail, Orthrus is real, peer reviewed, and traceable to two of the three founders. The business behind it, the named pharma users, the Arc Institute work, and the exact shape of the $7.2 million round, rests mostly on the company's own account so far.
What to watch
- Whether a Form D appears under Blank Bio's own name, which would confirm the size and terms of the seed round.
- Named pharma or diagnostic customers, beyond the unconfirmed reference to Sanofi and GSK using the open source model.
- Results from the PacBio tumor sequencing collaboration.
- Any published paper describing Blank Bio's production model, as opposed to the academic Orthrus work.
In their words
“Bulk RNA-seq is one of the most clinically accessible and information-rich assays in oncology, but much of its signal is still reduced to simplified gene-level summaries.”
Jonathan Hsu, CEO and co-founder, Business Wire release, May 19, 2026 · Company release
“Blank Bio was founded to apply foundation models to the full molecular detail contained in each patient’s tumor transcriptome and turn that information into more precise, clinically useful predictions.”
Jonathan Hsu, CEO and co-founder, Business Wire release, May 19, 2026 · Company release
“PacBio HiFi long-read sequencing was built to resolve biology that other technologies miss, and nowhere is that more consequential than in the complex transcriptomes of patient tumors.”
David Miller, Global Vice President of Marketing, PacBio, Business Wire release, May 19, 2026 · Partner stated
“Blank Bio is building at the intersection of two major shifts in biology: the expanding clinical use of RNA-seq and the emergence of foundation models capable of learning complex biological patterns at scale.”
Sahir Raoof, TechBio advisor, SignalFire, Business Wire release, May 19, 2026 · Investor stated
“mRNA enables programmable, personalized medicine, from gene therapies to cancer vaccines. But with more possible sequences than stars in the universe, brute-force design is impossible.”
Jonathan Hsu, Ruian Shi and Philip Fradkin, Y Combinator launch post, 2025 · Founders
“Just defended my PhD in CS at the University of Toronto. Published at ICML, NeurIPS, Nature Comms; used 4 GPUs to pre-train an mRNA foundation model that beat competitors using months of compute on thousands of H100s.”
Ruian Shi, co-founder and CTO, Y Combinator company page, 2025 · Founder
Related companies
Sources
- Public recordForm D, AVSF, Blank Bio 2025, LLC
- Public recordOrthrus: Towards Evolutionary and Functional RNA Foundation Models
- Public recordOrthrus: Towards Evolutionary and Functional RNA Foundation Models
- CompanyBlank Bio website
- CompanyBlank Bio company page and founder launch post
- CompanyBlank Bio Announces Seed Financing and Strategic Collaboration with PacBio to Advance RNA Foundation Models for Precision Oncology
- IndependentBlank Bio Announces Seed Financing and Strategic Collaboration with PacBio to Advance RNA Foundation Models for Precision Oncology
- IndependentBlank Bio: $7.2 Million Seed Financing Raised To Advance RNA Foundation Models For Precision Oncology
- IndependentBlank Bio Raises $7.2M in Seed Funding
- IndependentBlank Bio Founder Genealogy
- IndependentBlank Bio, 2026 Company Profile & Team
Profile researched and written by Healthcare Discovery. Last updated September 29, 2026.
