← All use cases

Annotate scRNA-seq data

Go through single-cell QC, annotations, and UMAPs within one conversation.

Use the NGS Analysis plugin with ZeroTwo to take a 10x-style matrix bundle through QC. You get filtered single-cell outputs, summaries that justify each threshold, annotations, and UMAPs, all open to inspection and revision in the same chat.

Intermediate
1h

Best for

  • Single-cell teams running QC, annotation, and plots after counting.
  • Auditable, threshold-justified filtering.

Skills & Plugins

  • NGS Analysis
    Routes the job to its post-count QC skill, with the scRNA-seq skill for FASTQ-to-count work.

Starter prompt

I've uploaded a 10x-style bundle (matrix.mtx, barcodes, features) plus a manifest and sample metadata for six mouse kidney samples. Run QC using the NGS Analysis plugin: choose filtering thresholds from the observed distributions and explain each choice, keep the raw counts intact, then annotate cell types and make UMAPs colored by label and by cluster. Send back a summary, the per-sample QC table, the justification plots, and a filtered .h5ad.

Leverage skills

Three packaged skills ship together: ngs-analysis-router, scrna-seq-qc, and ngs-scrna-seq, all available once the plugin is on. Invoke them from the composer with $, /, or @.

Step-by-step guide

  1. Supply inputs

    Matrix, barcodes, features, manifest, metadata.

  2. Run the prompt

    Thresholds and rationale get recorded.

  3. Inspect

    Open the generated visualizations/index.html, the review notebook, or the local review app it launches.

  4. Iterate

    Adjust thresholds, add a reference atlas, or rerun.

With no matched reference, the built-in marker fallback is tuned for blood cells (PBMC). For other tissues, supply a reference atlas or review the labels yourself.

Results

Start with the justification plots and QC summary, then the UMAPs by label and cluster, then cell-level metrics. Each run writes a summary and a status file that states what finished and what was blocked, so a revision can be compared with the first-pass rationale.

Related use cases