Analyze
Analyze runs computational workflows. Tasks are grouped by research intent rather than by software package name.
Overview
Shows task readiness (which steps have completed runs), recommended next actions, and links to stale outputs.
Find Structure
| Step | Purpose |
|---|---|
| QC | Filter cells by counts, genes, and mitochondrial percentage |
| Normalization | Log-normalize, scale, select highly variable genes |
| Clustering | Neighbors graph, Leiden clustering, UMAP |
| Spatial domains | Cluster spatial spots into tissue domains (spatial datasets only) |
Steps must generally run in order; each saves a checkpoint .h5ad for the next step.
Compare Groups
| Step | Purpose |
|---|---|
| Differential expression | Wilcoxon rank-sum on contrasts or cluster markers |
| Pathway enrichment | Over-representation analysis (ORA) on DE gene lists |
For contrast-based DE, save contrasts on the Data page first. Cluster marker mode compares each cluster to all others.
Build Predictors
| Step | Purpose |
|---|---|
| Biomarker discovery | Feature selection (mRMR or random forest), optional WGCNA, classifier CV |
Requires a class label column in obs (default condition) with at least two classes.
Integrate / Scale
Placeholder for future batch integration, reference mapping, and million-cell sketch workflows. Not yet available.
Result pages
Completed steps link to dedicated result views:
- DE Results — volcano plot and sortable gene table
- Biomarker Results — ranked genes and classifier metrics
- Enrichment opens from Interpret or Analyze links
Errors and reruns
Failed runs show the error message on Analyze and in expandable rows on Runs. Fix parameters or input data, then re-run. Re-running QC or normalization invalidates downstream checkpoints.