Scienta

🇺🇸 Scienta is heading to BIO 2026 in San Diego (June 22–25) — proud to be there as an EIC 2025 Laureate. We're bringing EVA, our foundation model for precision immunology, to the world's largest biotech gathering. Come find us: 📍 Booth 2411 — European Pavilion, Hall C 🗣️ “TechBio Without Borders” panel with Vincent Bouget — June 22, 3:00 PM · TechBio Booth, French Pavilion (organized by Business France North America) 🎤 Start-Up Stadium pitch — June 22, 4:00 PM · Room 5B (Level 2) 🤝 Book a meeting via BIO Partnering Camille Bouget, Vincent Bouget & Julien Duquesne will be on the ground — let's connect. Swipe through to meet EVA 👉 #EICatBIO2026 #EUeic #TechBio #BIO2026 #Immunology #AI #foundationmodel

Et si l’IA permettait enfin d’accélérer la compréhension… du système immunitaire ? Scienta franchit une étape ambitieuse avec EVA, son modèle monde dédié aux maladies auto-immunes et inflammatoires, qui simule le fonctionnement du système immunitaire pour mieux comprendre, prédire… et accélérer la découverte de traitements. Un sujet dès plus intéressant à la croisée de l’IA et de la biotech à découvrir dans LeMagIT Gaétan RAOUL. #IA #HealthTech #Biotech #Innovation #Immunologie https://lnkd.in/egPJRB9J

Developing new drugs in immunology & inflammation remains a translational challenge. - Which targets should be prioritized? - Which therapeutic indications are most likely to show efficacy? - Which patients will respond to treatment? Early in the process, human evidence is scarce and decisions rely heavily on preclinical models. Yet biological signal rarely translate into patient-level outcomes, as reflected in the high failure rates of clinical trials. EVA was developed to address this translational gap. EVA is a foundation model for immunology & inflammation that integrates multimodal biological data to generate patient-level predictions. Discover how EVA can support your pipeline, from target discovery to clinical Link to white paper available in comment

We're introducing EVA, a 440M-parameter cross-species, multimodal foundation model built specifically for immunology and inflammation. Drug development in immuno-inflammation still faces a major translation gap, the well-known valley of death between preclinical signals and clinical outcomes. This week marks an important milestone for the team as we present a significant advance in our AI platform. EVA integrates transcriptomics and histology across human and mouse into unified patient-level representations. Trained on 500K+ transcriptomics samples and 20M histology tiles across 50+ tissues and conditions, it learns what is shared across species and modalities. We evaluated EVA on nearly 40 tasks designed to mirror real R&D decisions across the drug discovery pipeline. It consistently outperforms existing foundation models, and critically, it doesn't lose to simple baselines. One result we're particularly excited about: across 6 immune-mediated diseases and 28 drugs, EVA's in silico predictions more than double the effective success rate compared to historical Phase II outcomes. We release an open version of EVA for transcriptomics data, to accelerate translational research in immunology. Technical report, blog post, and model weights in the comments. Camille Bouget Vincent Bouget Julien Duquesne

𝗥𝗲𝘁𝗼𝘂𝗿 𝘀𝘂𝗿 𝗹’𝗲́𝗱𝗶𝘁𝗶𝗼𝗻 𝟮𝟬𝟮𝟱 𝗱𝗲 𝗹𝗮 𝗰𝗼𝗻𝗳𝗲́𝗿𝗲𝗻𝗰𝗲 𝗧𝗲𝗰𝗵𝗕𝗶𝗼 𝗙𝗿𝗮𝗻𝗰𝗲 : 𝘂𝗻𝗲 𝗺𝗮𝘁𝗶𝗻𝗲́𝗲 𝗿𝗶𝗰𝗵𝗲 𝗲𝗻 𝗲́𝗰𝗵𝗮𝗻𝗴𝗲𝘀 𝗲𝘁 𝗲𝗻 𝗽𝗲𝗿𝘀𝗽𝗲𝗰𝘁𝗶𝘃𝗲𝘀 𝗽𝗼𝘂𝗿 𝗹𝗮 𝗳𝗶𝗹𝗶𝗲̀𝗿𝗲 ! Hier, la Commission TechBio, aux côtés de France Biotech et France Deeptech, et co-coordonnée par WhiteLab Genomics et Scienta Lab, organisait la seconde édition de TechBio France 2025, accueillie par Future4care. Une matinée qui confirme l’élan d’un secteur au croisement de l’IA, de la biologie et de la recherche translationnelle. Merci à France Biotech, Deep Francis, PhD, et l’équipe Future4care pour leur soutien et leur accueil. Merci également à Timothé CYNOBER (Pharm D.) pour son implication déterminante dans la co-coordination de la commission. 🎙️𝗧𝗲𝗺𝗽𝘀 𝗳𝗼𝗿𝘁𝘀 𝗱𝘂 𝗽𝗿𝗼𝗴𝗿𝗮𝗺𝗺𝗲 Trois tables rondes ont abordé des enjeux clés : financement non dilutif, propriété intellectuelle, création de valeur des plateformes TechBio et rôle croissant de l’IA dans la R&D pharmaceutique. Notre CEO Camille Bouget est intervenue dans la table ronde « Unlocking Non-Dilutive Funding » pour partager l’expérience de Scienta Lab en tant que lauréat EIC Accelerator 2025. 🗒️ 𝗗𝗲𝘂𝘅 𝗹𝗶𝘃𝗿𝗮𝗯𝗹𝗲𝘀 𝗽𝗿𝗲́𝘀𝗲𝗻𝘁𝗲́𝘀 :  - Speedinvest x DealFlow – Analyse des dynamiques d’investissement VC en TechBio. - Rapport TechBio 2025 – Synthèse des travaux de la commission et vision prospective de la filière: https://lnkd.in/eXgEjx42  🏛️ 𝗨𝗻𝗲 𝗳𝗶𝗹𝗶𝗲̀𝗿𝗲 𝗾𝘂𝗶 𝘀’𝗼𝗿𝗴𝗮𝗻𝗶𝘀𝗲 𝗲𝘁 𝗽𝗼𝗿𝘁𝗲 𝘂𝗻𝗲 𝗮𝗺𝗯𝗶𝘁𝗶𝗼𝗻 𝗰𝗼𝗺𝗺𝘂𝗻𝗲 : La Commission TechBio œuvre à : • représenter la filière auprès des décideurs nationaux et européens, • renforcer les liens entre startups, industriels, investisseurs et académiques, • formuler des recommandations clés sur les données biologiques, l’accès aux financements et l’usage de l’IA en santé, • créer les conditions d’émergence de champions européens de la TechBio. Merci à l’ensemble des intervenants, participants et partenaires. Une édition 2025 qui marque une nouvelle étape pour la communauté TechBio — et que Scienta Lab est fier de contribuer à structurer. Genopole | Nucleate | In Extenso | Frédéric Girard | Romain Roullois | Yannick Menel | David Del Bourgo | Timothé CYNOBER (Pharm D.) | Philippe Moingeon | Camille Bouget | Hela Ammar | Robin Eggert-Griscelli | Agnès De Leersnyder | Romane Dorado Doncel | Catherine Martre | Chloe Evans

Proud to share our latest research — soon to be presented as a poster at the ML4Molecules workshop at EurIPS on December 2! This work, led by Charlotte Claye, Pierre MARSCHALL, and Julien Duquesne from the Scienta Lab team, focuses on improving the interpretability of single-cell foundation models such as scVI and scGPT. Building on our previous work applying explainable AI to histology (including our Lancet Rheumatology publication), this study extends concept-based interpretability to single-cell models, applied here to large immune-cell datasets, a key area for immunology and inflammatory disease research. In this study, we introduce a framework to interpret biological concepts from  scRNA-seq foundation models including an immunologist expert in the loop, enabling a clearer view of the immune mechanisms these models learn. 🔍 A few key findings: • Concepts extracted with sparse auto-encoders are more interpretable than model neurons. • Several concepts correspond to biologically meaningful immune cell types and processes. • Some concepts show cross-dataset stability, suggesting robust underlying structure. • Concept-based interpretability preserves information and supports interpretable downstream tasks. This work opens new possibilities for transparent, expert-friendly single-cell AI models — with promising applications in biomarker identification, target discovery, and biological insight generation. 📄 Full preprint: https://lnkd.in/eYK7-zfx

New peer-reviewed publication in The Lancet Rheumatology ! We’re proud to share that our co-founders Vincent Bouget and Julien Duquesne, together with an international and interdisciplinary team of clinicians, pathologists, and AI experts, have co-authored a landmark study just published in The Lancet Rheumatology: “Machine learning to classify the focus score and Sjögren’s disease using digitalized salivary gland biopsies.” Diagnosing Sjögren’s disease remains a clinical challenge — assessing the focus score from salivary-gland biopsies is a complex and subjective process that can vary significantly between experts. Conducted across six European expert centres and including 545 participants, the study demonstrates how explainable deep learning can not only improve the consistency and interpretability of histological assessment, but also drive scientific discovery by revealing novel histological biomarkers. Key highlights: ✅ AUROC up to 0.92 for Sjögren’s-disease diagnosis in external validation ✅ Identification of a new histological pattern — CD8⁺ T-cell infiltration around acinarepithelial cells, providing new insights into the disease’s heterogeneity ✅ A white-box AI approach enhancing interpretability, reproducibility, and biomarker discovery. A collaborative work between Scienta Lab and The European NECESSITY Consortium, paving the way for more objective and reproducible diagnostics in autoimmune diseases. 👉 Discover the key takeaways in the carousel below. 📄 Read the full article: https://lnkd.in/ecAeidPp

Scienta Lab is attending BioTechX Europe 2025 in Basel, Switzerland 🚀. Meet Julien Duquesne (CTO & Co-Founder), Jonathan Plassais (Product Lead), and Francesco Strozzi (Lead Computational Biologist) from October 6–8. We’re excited to exchange ideas on how data-driven innovation is accelerating the future of precision medicine in pharma and healthcare.

Scienta Lab is heading to UEG - United European Gastroenterology Week 2025 ! Meet our Vincent Bouget Bouget (CSO & co-founder of Scienta Lab) and Marina Colombo, PhD (Data Partnerships Lead) from October 5 – 7 in Berlin. Looking forward to connecting with the gastroenterology community !