omicverse.single.gptcelltype_local

omicverse.single.gptcelltype_local(input, tissuename=None, speciename='human', model_name='Qwen/Qwen2-7B-Instruct', topgenenumber=10)[source]

Annotate cell types with a local instruction-tuned LLM.

Parameters:

  • input (dict or pandas.DataFrame) – Marker definition per cluster. Accepted formats: 1) dict[cluster_id -> list[str]] marker genes, or 2) DE table with cluster, names, logfoldchanges columns.

  • tissuename (str or None, default=None) – Tissue context included in the prompt.

  • speciename (str, default='human') – Species context string included in the prompt.

  • model_name (str, default='Qwen/Qwen2-7B-Instruct') – Local Hugging Face model name or path used for generation.

  • topgenenumber (int, default=10) – Maximum number of marker genes retained per cluster.

Returns:

Mapping from cluster ID to predicted cell type name.

Return type:

dict

Examples

>>> markers = {"0": ["CD3D", "IL7R"], "1": ["NKG7", "GNLY"]}
>>> res = gptcelltype_local(markers, tissuename="PBMC", speciename="human")