Every cell in your sample, named and explained.

It measures which genes are active in each individual cell of a sample, thousands to millions of cells at a time, instead of averaging over a whole tissue. The result is a table of counts: how many times each gene was read in each cell. That is what lets you see the cell types in a sample, how they change in disease, and where a treatment acts.

It takes that table of counts, before any processing, and characterises every cell in the sample, quantitatively and qualitatively, in minutes to hours rather than days or weeks, at a fraction of the usual cost. You receive a written report, a copy of your data with the findings attached to each cell, and an archive with every table and figure.

Because generated code still has to be reviewed and tested, and that work lands on you. A pipeline has to be designed, validated end to end, kept working as tools and references change, and run on the right hardware. We engineered it once, test every change, and it runs the same way for everyone, in one maintained container: nothing to install, nothing to verify each time. You also do not bring your own compute: the GPUs, the runtime and the scaling are ours to run, not yours to provision.

We stay close to current best practice, and the workflow evolves with feedback and as the field progresses. The methods page documents every step and its settings. The model weights are open, so anyone can reproduce a result outside our service, and each report records exactly what ran.

Two ways. If you have your own samples, the sequencing core or facility that processes them returns a count matrix, which scanpy saves as an AnnData (.h5ad) file in one line; ask for raw counts, not normalised values. If not, public archives: CELLxGENE hosts thousands of curated datasets as AnnData (.h5ad) files, free for research, and many studies deposit theirs in the Gene Expression Omnibus.

Yes. It runs on GPUs and has been tested on more than fifteen million cells in a single run, from a 59 GB file, a size at which typical notebook pipelines run out of memory or take days. If your dataset is larger than that, contact us first.

Yes. Results are yours to use in papers, grant applications, theses and internal work alike.

You do. You keep ownership of the data you upload and of every result the run produces: the terms claim no rights in either. We use your data only to run the job you submitted and to operate the service, and never to train anything.

Acknowledging VarnaOps is welcome but not required. Cite the original studies and the tools whose licences ask for it: what to cite, and how, is on the attribution page. If you use the weights outside the platform, their CC BY 4.0 licence asks you to credit VarnaOps, and the reference datasets carry CC BY 4.0 too.

Not for diagnosis or treatment decisions. It is a research tool, and that includes research done in hospitals and clinical labs. Using it in patient care would need a certified medical device, which this is not yet.

Your file is encrypted in transit and at rest, processed entirely inside our own Amazon Web Services (AWS) environment, and never used to train anything; we claim no rights to it or to your results. It is deleted seven days after upload and your results thirty days after the run, or sooner if you delete them yourself; the data handling page has the details.

A fixed fee per submission, currently a small fixed fee, plus a variable amount that scales with the computing your run needs: more cells, more genes, more donors and more analyses cost more. The price is shown before you start, and anything that could not run is refunded. Buy credits by card, or ask us for an invoice for your institution; $1 is 1 credit, and there is no subscription.