Single-cell atlas
Build a single-cell RNA-seq atlas from raw or processed data — sample-quality QC (doublets, ambient RNA, dissociation stress), clustering, scalable sketch analysis, batch integration, cell type annotation, marker identification, pseudobulk differential expression, composition testing, trajectory, RNA velocity, and pathway enrichment with reproducible history.
Research question
What cell populations exist in this dataset, what genes define each cluster, how do cell states change across trajectory or condition, and how does expression differ between experimental groups with biological replicates?
Who this is for
- Wet-lab biologists running their first single-cell analysis without local Seurat/Scanpy setup
- Computational biologists who want transparent parameters and checkpoint provenance
- Core facilities delivering standard QC → clustering → marker → enrichment packages to client labs
Data requirements
| Data | Required | Purpose |
|---|---|---|
.h5ad, 10x/CSV, or a Seurat / SingleCellExperiment .rds | Yes | AnnData input for all pipeline steps |
| Study organism | Recommended | Picks marker panels and GO/Reactome sets for human, mouse, or rat |
Spliced/unspliced counts (.loom) | No (required for RNA velocity) | Attached to the dataset on the Data page |
Gene symbols in var | Recommended | GO enrichment compatibility |
| Metadata with condition labels | No (required for contrast DE) | Group comparison on Data page |
| Saved contrasts | No (required for contrast DE) | Treated vs. control comparisons |
Workflow
Data → QC → (Ambient RNA) → Normalization → (Integration) → Clustering → Annotation / Trajectory / Velocity → DE or Pseudobulk DE / Composition → Enrichment → History snapshot
Step 1 — Upload and prepare
Create a single-cell study, set its organism, and upload an .h5ad file, 10x/CSV data, or a Seurat or SingleCellExperiment .rds (then click Import on the Data page). Optionally attach a raw droplet matrix (for ambient RNA correction) or velocity counts under the dataset. Save metadata on Data → Metadata so columns merge into obs before analysis runs.
Define contrasts on the Data page when condition-level DE is needed.
Step 2 — Find structure
Open Analyze → Find Structure and run in order:
- QC — filter by genes/cells detected, counts, and mitochondrial fraction; doublet scores, assay type, and stress signatures
- Ambient RNA Correction (optional) — remove cell-free RNA before normalization
- Normalization — log-normalize and select highly variable genes
- Clustering — Leiden clustering and UMAP embedding (run Batch Integration first for multi-sample data)
For large datasets, run Integrate / Scale → Sketch & Scale to create a representative sketch and projected full-dataset labels before deep exploratory work.
Re-running an upstream step invalidates downstream checkpoints and marks dependent results stale.
Step 3 — Explore
Review outputs before formal comparison:
- Explore → Sample QC — retention histograms, per-sample metrics, stress signatures, doublets, and ambient RNA
- Explore → Embedding — interactive UMAP colored by cluster or metadata column
Step 4 — Compare groups
Under Analyze → Compare Groups:
| Mode | When to use |
|---|---|
| Cluster markers | Genes enriched in each cluster vs. all others |
| Contrast | Groups defined on the Data page (e.g. treated vs. control) |
| Pseudobulk DE | Replicate-aware DE after aggregating cells by sample plus cell type or cluster |
| Cell Composition | Whether cell-type proportions change between conditions (needs replicates) |
| Preparation Method Comparison | How enzymatic, fixed-cell, and nuclei preparations differ in recovery, quality, composition, and markers |
Run Pathway enrichment on DE results with ORA for thresholded gene lists, ranked GSEA for full DE tables, or ssGSEA/GSVA for sample- or cell-level gene-set scores.
Step 5 — Integrate, annotate, and order cells
- Integrate / Scale → Batch Integration corrects batch effects with Harmony or fastMNN and reports batch mixing and cell-type preservation.
- Find Structure → Cell Type Annotation labels clusters from an organism-aware marker panel (blood, adrenal, pituitary, thyroid, testis), your own markers, or a labelled reference dataset, with scores, supporting genes, and marker sources.
- Find Structure → Trajectory & Cell Cycle scores S/G2M programs and computes graph-based pseudotime from a selected root cluster or cell.
- Find Structure → RNA Velocity shows where cells are heading from spliced/unspliced counts, with the spliced fraction per cell.
- Integrate / Scale → Sketch & Scale stores large-dataset Zarr/Parquet artifacts and downsampled browser payloads.
Step 6 — Interpret and preserve history
- Interpret → Enrichment — GO term tree and pathway context
- History → Snapshots — freeze parameter set and run IDs for reproducibility
Step 7 — AI interpretation (optional)
Use analytical interpretation on completed DE or enrichment runs for cluster annotations and pathway narratives. If Interpret isn't available, ask your administrator to enable AI interpretation.
Expected outputs
- Filtered AnnData checkpoint for the run
- UMAP embedding with Leiden cluster labels
- Doublet, ambient RNA, and stress-signature QC per sample
- Integrated PCA/UMAP-ready checkpoint with batch diagnostics
- Cell type annotation table with scores, confidence, and supporting markers
- Cell-cycle phase and pseudotime summaries; RNA velocity arrows and gene fits
- Composition tests by condition, and preparation-method comparisons
- Ranked DE table with log fold change and adjusted p-values
- Pseudobulk DE table by sample and cell type or cluster when biological replicates are available
- GO enrichment results with term hierarchy
- Analysis snapshot with completed run IDs and parameters
Typical analyses
| Analysis | Comparison | Question |
|---|---|---|
| Cell type discovery | Cluster markers | What genes define each population? |
| Treatment response | Contrast DE | Which genes change after perturbation? |
| Replicate-aware response | Pseudobulk DE | Which cell types show sample-level treatment effects? |
| Multi-sample atlas | Batch Integration + Annotation | Are cell labels stable after batch correction? |
| Differentiation or activation | Trajectory & Cell Cycle, RNA Velocity | How are cell states ordered, and where are they heading? |
| Population shifts | Cell Composition | Is a cell type expanded or depleted in the knockout? |
| Protocol choice | Preparation Method Comparison | Does a fixed-cell or nuclei protocol preserve cell types and markers as well as fresh dissociation? |
| Immune infiltration | Condition on metadata | Are T cell clusters expanded in responders? |
| Core client delivery | Full pipeline + snapshot | Reproducible deliverable for requesting lab |