Observed arrival · 2026-11-17
Cell Atlas puts uncertainty in the annotation loop
A browser-based research prototype annotates single-cell expression matrices against fitted references, with confidence gates, broad-to-fine labels, and diagnostics.
- For
- Researchers exploring single-cell expression data
- Worth noticing
- It caps uploads at 200 cells and warns that aligned gene identifiers alone did not prevent high-error cross-study transfer.
Field notes
Users can provide a CSV matrix (cells by named genes) or a JSON object for one cell; the page caps uploads at 200 cells and says missing values are not guessed. Inference runs in a browser worker, with only public model assets served by the host, and results can be exported as CSV or diagnostics JSON. The follow-up reports seven primary references and explicitly says its agreement measures are not independent biological validation.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue