Summary
- Rivercell has raised a $25 million seed round led by HV with HCVC, Alven and Bpifrance Digital Venture participating.
- Funding will expand its Paris wet lab, proprietary cellular data generation and AI Virtual Cell programme.
- The company has not disclosed the dataset size or prediction benchmarks required to assess its modelling advantage.
Rivercell has raised $25 million to build both the experimental data infrastructure and AI models behind a proposed virtual cell, arguing that better biological prediction depends first on generating datasets that existing research infrastructure does not provide.
The Paris company emerged from stealth on 7 October with a seed round led by HV and joined by HCVC, Alven and Bpifrance Digital Venture. The capital will expand Rivercell’s wet lab, scale its proprietary data-generation platform and support an AI Virtual Cell programme intended to predict how human cells respond to drugs and genetic interventions.
Many computational drug discovery businesses begin with biological datasets collected for other scientific purposes and then apply machine learning to identify patterns or candidates. Rivercell is taking a more integrated approach by designing experiments specifically to produce training material for its own model.
The company plans to generate interventional, time-resolved and multimodal single-cell data. Cells are exposed to chemical or genetic changes and measured through multiple biological signals over time, giving a model information about how a response develops rather than recording only a static before-and-after comparison.
The bottleneck sits in experimental data
Rivercell’s strategy assumes that increasingly capable cellular models cannot emerge from computing scale alone because structured examples of how cells respond to interventions remain comparatively scarce. Existing biological datasets are extensive, but they were not necessarily designed around the causal and temporal relationships a predictive model needs to learn.
Generating those observations requires laboratory throughput alongside machine learning. Rivercell is developing proprietary equipment and expanding an automated wet lab in Paris so it can perform many interventions, observe the resulting cellular responses and return those measurements to its training pipeline.
The company compares that infrastructure with the computing layer used to train large AI models, although biology introduces additional constraints. Living systems contain variability, experiments can be difficult to standardise and measurement error can become part of the data, meaning scale only helps when protocols remain consistent enough for a model to learn biological relationships rather than laboratory artefacts.
Automation can reduce some of that variability while increasing the number of experiments that can be run, but the value of the resulting dataset will depend on experimental design and coverage as much as its overall size.
A model still has to prove predictive value
Rivercell ultimately wants its virtual cell to predict the effect of a drug or genetic change before researchers run every experiment physically. A sufficiently reliable model could narrow the range of candidate interventions that need to proceed through expensive laboratory work.
Physical validation would still remain necessary, especially as a programme moves towards a therapeutic candidate. Cell types, disease states and experimental conditions can alter biological behaviour, while predictions developed in isolated cells do not automatically transfer to a complete organism.
Rivercell says the model will be indication agnostic, learning more general rules of cellular response rather than being built for a single disease. The company points to possible applications across oncology, immunology, rare diseases and cardiometabolic conditions, substantially widening the range of biology the model would eventually need to represent.
The company has not yet disclosed the size of its proprietary dataset, quantitative prediction benchmarks or a release date for the virtual cell. Those missing figures prevent an external assessment of whether the integrated data-generation strategy currently gives Rivercell a measurable modelling advantage.
The business was founded in 2025 by Yann Fleureau, previously co-founder and chief executive of cardiac diagnostics company Cardiologs, alongside Eric Durand, who has held senior data-science roles at Novartis and Owkin and co-founded biological foundation-model company Bioptimus.
Competition in virtual cells and biological foundation models is increasing, which means proprietary experimental infrastructure may become as significant as model architecture. Companies able to design the data they need can potentially train on observations competitors cannot obtain from public repositories.
Rivercell’s $25 million round is therefore financing a biological data operation as much as an AI model. The useful measure of progress will be whether experiments produced through that infrastructure lead to predictions accurate enough to change which drug-development decisions researchers make next.
