Cell Studio
Single-cell RNA-seq analysis, without writing R.
Drop in a count matrix and run the full single-cell pipeline in your browser: quality control, normalization, variable genes, PCA, batch correction, clustering, UMAP and t-SNE, marker genes, and Wilcoxon differential expression. Everything runs on your own computer with WebAssembly, so your data is never uploaded anywhere.
Start with the demo: the pancreatic cancer data behind Ridhi's single-cell project (6,618 mouse cells from Burdziak et al., Science 2023), then check the Reproduce tab to see the browser match a scanpy analysis gene for gene.
Each group is compared with every other group; effect sizes are summarized across those pairwise comparisons. AUC is the probability that a cell in the group expresses the gene more than a cell outside it (0.5 = no difference).
Two-sided Wilcoxon rank-sum test with tie correction and Benjamini-Hochberg adjustment, on library-size-normalized counts. Fold changes use log1p(counts per 10,000) means, the same definition scanpy reports. Cells are treated as independent samples, so p values describe cells, not animals or patients.
This tab checks the browser against an independent analysis. The same 6,618 cells were analysed in Python with scanpy (Wilcoxon rank-sum, tie-corrected, Benjamini-Hochberg). Press the button to run the matching comparison here, all K5 PDAC cells against all K4 PanIN cells, and compare the ten genes from her study line by line.
If both columns agree, the browser tool reproduces the scanpy result from the same raw counts, so the same findings can be explored here without R or Python. The left panel compares the clusters found here with the cluster labels deposited by the authors, using the adjusted Rand index (1 = identical, 0 = chance). Different methods and settings move cluster boundaries, so a perfect score is not expected.
Plots export from their own panels as PNG or SVG.
The scripts follow the same steps with the closest equivalent settings. Seurat and scanpy implement normalization, variable gene selection and clustering differently from scran.js, so expect similar but not identical clusters.
How it works
Single-cell analysis usually means an R or Python environment, a package install, and a script. Cell Studio runs the same established methods directly in the browser, compiled to WebAssembly from the C++ libraries behind the Bioconductor package scran, through scran.js. The work runs on several CPU threads in the background, so the page stays responsive.
Private by design
Files are read on your machine. Nothing is uploaded, so unpublished data stay on your computer.
Every step adjustable
Change a setting and only the steps after it re-run. Clusters, UMAP and markers update in seconds on a laptop.
Built to sit beside a paper
Readers can open the dataset, recolor the map by any gene, and rerun the comparisons, with a methods paragraph and R and Python scripts to match.