Robust and efficient annotation of cell states through gene signature scoring

Table 1.

Qualitative comparison of signature scoring methods based on their properties, influencing factors, and score comparability

Scoring method Deterministic Optimal control gene selection Independent of total cell number and cell-type proportions Robust to batch effect when scoring all samples together Robust when scoring for small signatures Robust to unrelated genes in the signature Provides comparable scores for multiple cell-type signatures
Adjusted Neighborhood Scoring (ANS)
SCANPY (Wolf et al. 2018): based on work by Tirosh et al. (2016)
Seurat (Satija et al. 2015): based on work by Tirosh et al. (2016)
Seurat_AG
Seurat_LVG
Jasmine_LH (Noureen et al. 2022) NA
Jasmine_OR (Noureen et al. 2022) NA
UCell (Andreatta and Carmona 2021) NA
  • A check mark indicates a scoring method fulfills property or resists an influencing factor; a cross mark, the opposite. (Seurat_AG) Seurat scoring with all genes as control; (Seurat_LVG) Seurat scoring with least variable genes as control; (Jasmine_LH) JASMINE with likelihood computation; (Jasmine_OR) JASMINE with odds-ratio computation; (NA) a property or an influencing factor that cannot be evaluated for a scoring method. These results are discussed in detail in the Results section.

This Article

  1. Genome Res. 36: 630-644

Preprint Server