In a lab in Villejuif, a slice of a patient's colon tumor is still dividing. Orakl Oncology grows these fragments into organoids, screens cancer drugs against them, and pairs each response with the patient's clinical history. The company calls the result a virtual patient.

On 12 November, Orakl raised a $34 million Series A to build more of them.

The pitch contains a paradox. Orakl sells foresight: a forecast of how a Phase 2 trial will end before it enrolls anyone. The evidence behind that claim comes from trials that have already finished.

What the round actually buys

Norrsken VC and Singular co-led the round, described as oversubscribed. InfraVia, Serena, Bpifrance's Patient Autonome Fund, teampact.ventures and Plug and Play joined; Verve Ventures, Speedinvest and HCVC returned. Total funding is now $50 million, after a €3 million round in 2023 and €11 million in 2024.

Orakl spun out of Gustave Roussy in 2023. It has more than 40 staff, 30% of them PhDs. CEO Fanny Jaulin spent two decades in cancer biology. CTO Gustave Ronteix works on AI and quantitative biology. COO Diane-Laure Pagès runs hospital partnerships and operations.

The money is earmarked for fivefold more experimental data, expansion from colorectal and pancreatic cancer into non-small cell lung and gastric cancer, teams in the US and UK, and a new target-discovery effort.

The round buys data volume and hospital access. The organoids and the clinical records attached to them are the scarce asset, not the model weights.

The number the deck leads with

The company says its engine predicted cancer trial outcomes "prospectively with 90% accuracy." The supporting paper is less dramatic. SCOPE, posted to medRxiv in April 2026 and not peer reviewed, was built on 54 treatment lines from 52 patients and a biobank of 81 organoid lines. It was tested against 32 arms from 23 published colorectal and pancreatic trials.

Median progression-free survival forecasts landed within 0.82 months of the reported result, with an R² of 0.85 and a Pearson correlation of 0.92. Objective response rate predictions were weaker: an R² of 0.71 and a 7.3 percentage-point error. The model picked the better arm in seven of eight head-to-head comparisons.

📊
What the preprint measured

32 arms from 23 published trials · median PFS error 0.82 months · R² 0.85 · better arm in 7 of 8 · 88% of arms conventional chemotherapy

Then the caveat that decides what the number is worth. Conventional chemotherapy accounted for 88% of the arms. The validation ran backwards, against trials whose results were already public. FirstWord HealthTech asked how the 90% was measured. The company did not provide details.

Orakl points to daraxonrasib, a pan-RAS inhibitor, where the preprint's forecast for the KRAS-mutant subgroup was consistent with later trial data. That is a useful anecdote. It is not a prospective trial of a new mechanism.

For decades, we've developed cancer drugs without being able to reliably predict how they will behave in patients until they reach clinical trials.— Fanny Jaulin, CEO and co-founder, Orakl Oncology

Who pays for a trial forecast

Roughly 75% of oncology candidates fail in Phase 2, according to BIO and Informa data covering 2011–2020. A forecast that changes one go/no-go decision is worth more than its price to a large drugmaker. That arithmetic, not the accuracy claim, is what the round is underwriting.

Orakl says 18 of the world's 20 largest pharma companies are in its commercial pipeline, with more than 10 drug programs and more than 15 additional cancer centers in contract talks. In July it secured exclusive rights to evaluate four undisclosed Sanofi oncology assets. No pharma partner is named, and no prediction has been published as a decision outcome.

The competitive frame is broader than AI drug discovery. Organ-on-chip vendors, digital-twin builders and contract research organizations all sell a better proxy for the patient. Orakl's differentiator is the closed loop: organoid screen, clinical record, forecast, then a lab test of the target.

We have seen this movie in clinical-trial software. As we wrote in November, the bottleneck is usually the system of record, not the model. Orakl's bet is that owning the biological data makes it the system of record for trial design.

The next real data point will not be another round or another preprint. It will be a pharma partner disclosing that it killed or advanced a program because an Orakl forecast said so. Until then, the virtual patients are a promising instrument, and the $34 million is a wager that enough of them will make the difference.

Orakl Oncology raises $34 million to make cancer drug response predictable
The primary announcement: round structure, investor list, use of funds, founders and the 90% accuracy claim.
Read the claim in the company's own words before weighing the independent reporting.
Grown, not scraped: the case for and against patient avatars
A detailed reading of the SCOPE preprint, its trial counts and error bars, plus the commercial and regulatory context.
The most specific public breakdown of what the model actually predicted.
Orakl raises $34M to scale AI platform predicting cancer drug responses
Trade coverage that pairs the funding with a direct reality check on retrospective validation and the chemotherapy-heavy test set.
The caveats that the funding announcement leaves out.