omicverse-single-cell-annotation

To annotate single-cell AnnData, use a reusable workflow that maps clustered data to cell-type labels via CellTypist, GPT-based, or SCSA backends. It writes branch-specific label columns and prediction matrices directly into the AnnData object for downstream analysis.

Mapping leiden clusters to cell types without manual curation is achieved by using automated annotation backends like CellTypist for pretrained model predictions, gpt4celltype for LLM-driven marker-to-label mapping, or SCSA for database-scoring annotation.

Yes, single-cell annotation requires an annotation-ready AnnData object that is preprocessed and clustered. The AnnData must contain a cluster column, and you need backend-specific resources like a CellTypist model file, local SCSA database, or AGI_API_KEY for LLM providers.

Yes, single-cell annotation supports multiple backends including CellTypist for pretrained model predictions, gpt4celltype for LLM-driven marker-to-label mapping, and SCSA for marker-database scoring, allowing backend-selectable annotation workflows.

LLM-based single-cell annotation limitations include the requirement of an AGI_API_KEY for LLM providers and the necessity of an annotation-ready AnnData with a pre-existing cluster column, as validation checks enforce these backend-specific resources before processing.

Yes, you can use gpt4celltype for marker-to-label mapping in single-cell RNA-seq workflows. It serves as an LLM-driven backend within the annotation workflow, generating prompts to map clustered AnnData data to reliable cell-type labels.