Oncology diagnostics · Functional precision medicine

Known Medicine

A Salt Lake City company that grew hundreds of micro-tumors from one patient sample and dosed each, to tell an oncologist what would work. It raised $7.2 million, registered a 2,000 patient study, published no validation data, and was absorbed into Pathos AI in 2023.

Three hundred tumors from one

A surgeon takes out at least a cubic centimetre of tumor, or drains more than 100 millilitres of fluid from around a patient’s lung. Under the protocol Known Medicine registered, that tissue had to be moving toward Salt Lake City within 24 hours, in a kit the company mailed out, with a vial of the patient’s blood.

In the lab the tissue was washed and digested to single cells. Cancer cells were sorted from stromal cells, stained with a membrane dye and set into a 3D matrix in a multi-well plate. The cultures rested 48 hours. Then the drugs went in. Seventy two hours later a fluorescence camera read every well.

From one sample the company said it could build more than 300 micro-tumors and test more than 30 treatments, with the readout reaching the oncologist inside seven days, weeks before a sequencing report. That is the whole proposition: run the experiment on the tumor instead of on the patient.

The record this idea had to walk past

The idea is old and the evidence behind it has been unkind for a long time. Testing a patient’s own cancer cells against drugs in a dish has a name in the literature, the chemotherapy sensitivity and resistance assay. In 2004 the American Society of Clinical Oncology reviewed the field and advised against using it outside a trial, citing small samples, few prospective studies and low assay yield.

In 2011 an ASCO working group went back through everything published between December 2003 and May 2010. Of 11,313 articles, five randomised trials survived to data extraction. The conclusion was blunt. “Review of the literature does not identify any CSRAs for which the evidence base is sufficient to support use in oncology practice.” The recommendation followed: “The use of CSRAs to select chemotherapeutic agents for individual patients is not recommended outside of the clinical trial setting.” ASCO said it again in a 2014 breast cancer guideline: “in vitro chemoresistance assays should not be used to select treatment.”

Known Medicine’s answer was that the biology underneath had changed. The old assays grew cancer cells alone on flat plastic. Its M3DUSA model put cancer, stromal and immune cells together in a hydrogel scaffold built for robots, and read the result with machine learning on images rather than a live and dead count. Whether that is enough is an empirical question.

A thesis defence and a shutdown

Andrea Mazzocchi did a PhD in biomedical engineering at Wake Forest, in the institute that had built tumor organoids carrying the patient’s own immune cells. Katie-Rose Skelly studied computer science, took a master’s in biomedical informatics and went to Utah as a data scientist at Recursion. They met through a mutual friend there.

They applied to Y Combinator in March 2020. “I defended my thesis two days before I had to apply to Y Combinator, the day North Carolina shut down for coronavirus. It was the wildest three days,” Mazzocchi told Fierce Biotech that August.

Y Combinator put in $150,000. Altitude Lab, the incubator Recursion runs with the University of Utah, gave them free bench space, and they outgrew it. By early 2021 the team was four and had moved into Research Park. The company says it closed 2021 with 13 people, hundreds of patient samples processed and millions of images.

The round, then the exit

On August 25, 2021 Known Medicine announced a $7.2 million seed round led by Caffeinated Capital, with Khosla Ventures, Cota Capital, Kickstart, Forward VC, OATV and Y Combinator, plus angels Chris Gibson of Recursion and Nish Bhat of Color.

In April 2022 it posted its validation study to ClinicalTrials.gov: KM3D, up to 2,000 patients, with Mayo Clinic, Duke Cancer Institute and MD Anderson as collaborators and Mazzocchi as principal investigator.

In March 2023 Skelly told Utah Business the team was working toward its first peer-reviewed publication, the paper that would be proof of concept. Two months later, on May 23, 2023, the company announced it had joined Pathos AI, started out of Tempus by Eric Lefkofsky and Ryan Fukushima. Fukushima wrote that day that Pathos had “also acquired and integrated Known Medicine, a company that builds patient-specific 3D cell cultures & uses AI approaches to prospectively predict drug responses in patients.” No terms were disclosed. The paper never appeared.

What is proven, and what is still claimed

Read plainly: Known Medicine built a real laboratory process, ran real patient samples through it, and signed up three serious cancer centres to feed it tissue. What it never produced, in public, is the number that matters. No paper, no posted result and no accuracy figure shows how often a prediction matched what happened to the patient. The study it registered asked that question retrospectively. The stronger question, whether patients treated by the assay do better than patients treated the usual way, was never put on the record here.

What to watch

• Whether any KM3D data is ever published. The registry entry has not moved since April 2023 and carries no results.
• Whether Pathos AI names the organoid platform in its own scientific or regulatory disclosures. Its 2026 platform page does not.
• Whether NCT05338073 is ever updated, completed or withdrawn.
• Whether anyone in functional precision oncology runs the randomised trial ASCO has asked for since 2004. That, not another platform launch, would move the guideline.

In their words

Known Medicine funding profile on HVCFHealthcare Venture Capital Fund

Related companies

• Correlia Biosystems: Similar focus. A UC Berkeley spinout that stopped waiting for proteins to drift toward an antibody and started pushing them there with an electric field
• Harmonic Discovery: Similar focus. A New York company using machine learning to design kinase drugs that engage several targets on purpose while avoiding the ones that cause harm
• Mecha Health: Similar focus. Four University College London researchers built a chest x-ray model that writes the whole report, published benchmarks saying it matches or beats radiologists, and
• Anja Health: Similar focus. A private cord blood, cord tissue and placenta bank built on TikTok, grown to about 2,000 families and $1.5 million a year, and sold in 2026

Sources

1. Public recordClinicalTrials.gov NCT05338073, KM3D Multicenter Cancer ConsortiumLast update Apr 24, 2023
2. Public recordASCO guideline update on chemotherapy sensitivity and resistance assaysJ Clin Oncol · 2011, PMID 21788567
3. Public recordASCO Technology Assessment on the same assaysNewswise · Jul 27, 2004
4. Public recordASCO guideline, advanced HER2-negative breast cancerThe ASCO Post · Mar 9, 2014
5. Public recordPubMed, NIH RePORTER and SEC EDGAR searchesRetrieved Sep 23, 2026
6. IndependentY Combinator startup grows mini-tumors to take the guesswork out of cancer treatmentsFierce Biotech · Aug 11, 2020
7. IndependentKnown Medicine is using machine learning to cure cancerUtah Business · Mar 28, 2023
8. IndependentKnown Medicine Closes $7.2M Seed Funding RoundPrecision Medicine Online · Aug 25, 2021
9. IndependentRe-engineering drug development for Precision MedicineRyan Fukushima, Pathos, on Medium · May 23, 2023
10. IndependentPathos AI Secures $365 Million in Series D FinancingBioUtah · May 15, 2025
11. IndependentChanging Cancer Care, One Organoid at a TimeWake Forest Baptist newsroom · Jan 21, 2020
12. CompanyKnown Medicine Raises $7.2 Million Seed RoundBusiness Wire · Aug 25, 2021
13. CompanyKnown Medicine story, application and home pagesknownmed.com · Sep 23, 2026
14. CompanyPosts on the Pathos acquisitionX · May 23, 2023
15. CompanyOur Platform, Pathos AIpathos.com · Sep 23, 2026