CHETAH

CHETAH classifies single-cell RNA sequencing data by hierarchical clustering of reference data to create a classification tree that enables step-wise, top-to-bottom classification with intermediate or unassigned categories.

Key Features:

• Hierarchical classification tree: CHETAH creates a classification tree by hierarchical clustering of reference data, enabling step-wise, top-to-bottom classification of input cells.
• Novel stopping rule: A stopping rule classifies input cells to reference cell types or to "intermediate types" representing more general classifications that ended in an intermediate node of the tree.
• Confidence scoring: Evidence for assignment includes a confidence score based on the variance in gene expression per cell type.
• Selective classification: CHETAH includes the possibility of intermediate or unassigned categories, providing superior specificity when cells of an unknown type are encountered.

Scientific Applications:

• Tumor sample analysis: CHETAH pinpoints malignant cells as intermediate or unassigned when they represent unknown cell types not present in reference data.
• Exploratory tissue studies: The use of unassigned and intermediate types highlights cell populations not well represented in reference datasets, including cells with profiles on a continuum between known cell types such as acinar and ductal cells in pancreas datasets.
• Prevention of misclassification: The possibility of unassigned and intermediate cell types prevents misclassification and yields important biological information for previously unexplored tissues.

Methodology:

CHETAH builds a classification tree through hierarchical clustering of reference data, then applies a novel stopping rule for step-wise, top-to-bottom classification of input cells based on variance in gene expression per cell type.