Scienta Lab and the EIC Accelerator: An AI Model of Immune Disease

Seven in ten immunology drug candidates fail to translate into benefit for patients, according to Scienta Lab. The CORDIS summary of Scienta Lab's EIC project says nearly half of clinical trial failures in immunology and inflammation are attributed to insufficient efficacy, and puts this down to preclinical models that represent human biology poorly. Scienta Lab, a French company founded in 2021, is building what it calls a foundation model of the immune system, EVA, to predict before a trial which targets, drugs and patients are likely to work.

The EIC Accelerator selected the project, PRISM (Predictive Research for Inflammatory Systems Medicine), at its March 2025 cut-off for blended finance. CORDIS records a grant agreement signed on 12 November 2025 with an EU contribution of €2,499,999, running from 1 December 2025 to 30 November 2027. Eleven months into the grant period, the model exists and has been described in a public preprint; whether it predicts a trial outcome that was not yet known has not been shown.

Three founders and a confidential balance sheet

The BODACC notices for the company show it was created on 11 June 2021 with €3,000 of capital and its seat in Balma, Haute-Garonne, and re-registered at the Évry commercial court in September 2021 after moving to 3 rue Joliot-Curie in Gif-sur-Yvette, on the Paris-Saclay campus. Vincent Bouget was the first president, with Julien Duquesne and Camille Bouget as directeurs généraux. A Université Paris-Saclay profile of January 2024 describes Duquesne as technology director, Vincent Bouget as scientific director and Camille Bouget, a pharmacist, as CEO, and puts the team at about ten people.

The register shows Camille Bouget as president from July 2023. A BODACC notice of 2 February 2026 records a change of management with Origae B.V., a company with the Dutch legal form B.V., as president, alongside the two directeurs généraux. A web search on 28 October 2026 found no public information on who owns or runs Origae B.V. The French state's company directory lists Scienta Lab as active, in the 10 to 19 employee bracket, with no published financial figures.

That last point matters for any assessment. Scienta Lab filed its accounts for 2022, 2023 and 2024, and each time BODACC records them as filed "accompagnés d'une déclaration de confidentialité", the option French small companies have to keep their accounts out of public view. Revenue, losses and cash are therefore not public.

What EVA is, according to the preprint

In February 2026 an eleven-author "Scienta Team" posted EVA: Towards a universal model of the immune system on arXiv. It describes a 440-million-parameter model with three parts: a gene expression encoder, a histology encoder and a fusion head. The expression side was pretrained on 545,343 bulk and single-cell transcriptomic samples from humans and mice, harmonised across platforms; the histology side on about 20.4 million image tiles from 4,076 whole-slide images of 1,252 patients. The authors test it on 39 tasks covering target discovery, cross-species translation of preclinical results and clinical questions such as treatment response and patient stratification, and report the best results in each category against models such as scGPT and BulkRNABert.

What sets EVA apart, as its authors present it, is the training data: it is built on immune and inflammatory disease specifically, and on both humans and mice, and the benchmark includes tasks that carry a perturbation measured in one species or disease over to another. The company's website says it covers more than 60 immuno-inflammatory diseases and 70 tissues, and that client data is used only to fine-tune a private copy of the model.

A version is open. The EVA-RNA model on Hugging Face has 48.6 million parameters, works on human and mouse gene expression, and is released under a company licence that requires users to register. The preprint's headline results come from the full 440-million-parameter system; the preprint describes the open release as covering transcriptomics only.

The number that matters, and what it rests on

For a drug developer, the question is whether EVA picks winners. The preprint's answer is a positive predictive value of 58% when it scores drug and disease pairs, set against what it gives as a 30% success rate for phase II trials in immunology and inflammation. The evaluation covers 28 drugs across six diseases, and the outcomes it is scored against are trials that have already reported. The paper describes no prospective test, in which the model's prediction is published before the trial reads out.

The company's own figures go further than the paper. Its website advertises "65% accuracy of clinical trial success predictions" and says EVA "doubles clinical trial success rates"; the preprint's abstract makes no such claim, and its own figure is 58% against a 30% base rate. The website also gives two different sizes for its training atlas, more than 200,000 profiles in one place and more than 96,000 in another. The EIC's project page carries the company's statements that it has published 20 peer-reviewed papers and has $7.5 million of total backing including the grant; neither figure is broken down there.

One named partner

On 29 January 2025 OSE Immunotherapeutics, a biotech based in Nantes and Paris and quoted on Euronext, announced a strategic partnership under which Scienta Lab would apply EVA to find signatures predicting which patients respond to OSE's immunotherapies in immune-mediated inflammatory diseases. HIT Consultant reported the same scope and no financial terms. No result from that collaboration has been announced. The website names OSE and carries a testimonial from Dennis McGonagle, a rheumatology professor at Leeds; it names no other biopharma customer.

The company is also a partner in AutoPiX, a €21.2 million Innovative Health Initiative project coordinated by the Medical University of Vienna to turn joint imaging in arthritis into quantitative biomarkers, running from November 2024 to October 2029. CORDIS records Scienta Lab's share as €330,875.

Money: a seed round and two EU grants

Scienta Lab announced its first round, €4 million, in December 2023, from CentraleSupélec Venture and business angels, per Station F and FrenchWeb, which names the angels Flavien Kulawik and Guillaume Amblard. BODACC records two capital increases before that announcement, in November 2022 and September 2023, which show new shares being issued but not what was paid for them.

The EIC selection was for blended finance, meaning a grant plus a possible equity investment by the EIC Fund. CORDIS records only the grant. The EIC website has a page for the company in its EIC Fund portfolio section, which returned "Access denied" when checked on 28 October 2026; no equity amount or date has been published by the EIC Fund or the company. No later private round could be found in company announcements or the press.

What PRISM promised and what can be checked

The CORDIS objective is to build a proprietary multimodal foundation model for immunology and inflammation that improves preclinical testing of new drug candidates, raises the success rate of clinical trials and supports precision medicine. The CORDIS record was last updated on 18 November 2025, before the project began, and lists no reports, deliverables or results yet, which is normal at this stage.

1. Multimodal foundation modelPublishedTranscriptomics and histology combined, described in the February 2026 preprint; whether it is the version funded by PRISM is not stated.
2. Better preclinical translationBenchmark onlyCross-species tasks reported in the preprint, not yet in a peer-reviewed journal.
3. Higher clinical trial successNot shownScored retrospectively on 28 drugs; no prospective prediction published.
4. Use in partners' drug programmesNot shownOSE partnership announced; no outcome reported.

In early 2026 the company presented work on predicting the clinical efficacy of inflammatory bowel disease candidates from preclinical data at the ECCO congress in Stockholm, per its website, and the same site lists the BIO International Convention of June 2026.

Why the EU funds it

The European Commission's Pharmaceutical Strategy for Europe of November 2020 counts the competitiveness and innovation of the EU pharmaceutical industry among its four pillars, and the Apply AI Strategy names healthcare and pharmaceuticals among the sectors where the EU wants AI adopted. A European model trained on immune disease data and sold to biotechs fits both policies, provided it shows that it can do what is claimed.

Scienta Lab was not asked for comment for this article. The reporting is based on publicly available sources, listed below.