Use this page when you need to interpret confidence, ambiguity, and edge cases in a CyteType report. For setup, privacy, pricing, and troubleshooting questions, see the FAQ.
Where CyteType is strong
CyteType is built for cluster-level annotation with ontology mapping, marker-level evidence, and a reviewable report. It tends to help most when:
- You already have solid clustering and marker gene priors
- You provide a specific
study_context(tissue, disease, organism, experiment) - You use confidence, heterogeneity, and reviewer alternatives to decide what needs human review
Confidence and review habits
Scores above 0.8 generally indicate high reliability for cell type, subtype, or state calls. Lower scores mark ambiguous or poorly supported populations that warrant manual review.
When a call looks surprising:
- Check marker-level evidence for unexpected or missing genes
- Read decision traceability for close runner-up candidates
- Use multi-expert synthesis alternatives as hypotheses, not errors
- Ask the Cluster Copilot before escalating
💡 Tip: High heterogeneity often means the cluster should be split or cleaned before you trust a subtype call. See Confidence and heterogeneity QC.
Known limits
- Annotations are assigned at the cluster level and propagated to cells
- Repeated runs can differ in wording while converging on stable biology; archive
query.jsonwhen you need a reproducibility record - UMAP display sampling does not change annotation inputs
- CyteType currently expects transcriptomic input; multimodal, CITE-seq, and spatial support are limited or in development
- The default cloud deployment should not be used with identifiable patient data without explicit data processing agreements
❗️ Important: For regulated environments, contact contact@nygen.io about enterprise options such as on-premises deployment, customer-managed LLMs, and zero data retention.
How to cite
Please cite the bioRxiv preprint: Ahuja G et al., Multi-agent AI enables evidence-based cell annotation in single-cell transcriptomics, bioRxiv 2025. doi:10.1101/2025.11.06.686964
Learn more
