Table 2 Strengths and weaknesses of immune cells quantification algorithms
Algorithm | Category | Strengths | Weakness |
|---|---|---|---|
ESTIMATE | G | Available for tumor purity and global immune status | Only a stromal score and an immune score are output. The information is limited27 |
xCell | G | Available for inference of 64 immune and stromal cell | The definitions of the cell subtypes are sometimes not clear Accuracy of prediction of some cell types is uncertain26 |
MCP-counter | G | Available for inference of fibroblasts and endothelial cells Available for an absolute quantification of specific cell population across samples Available for between-sample comparison | Relatively less cell types included in the inference (8 types) |
CIBERSORT | D | Available for inference of 22 immune cell subtypes Available for between-cell-type comparison | Relative proportion of distinct cell types in a single sample Trained on microarray rather than RNA-seq data30 |
EPIC | D | Available for inference of fibroblasts, endothelial cells, and uncharacterized cells Enabling inference of tumor purity from uncharacterized cell proportion Available for both between-sample and between-cell-type comparison | Only 6 immune cell types available Not available for discrimination of cell types with transcriptional similarity |
quanTIseq | D | Available for inference of 10 immune cell subtypes Available for both between-sample and between-cell-type comparison | Not available for quantification of stromal cells (e.g., cancer-associated fibroblasts) |
TIMER | D | A user-friendly analytic web tool for cancer immunology research | Only 6 immune cell types and no stromal cells available Relative proportion of distinct cell types in a single sample |
CIBERSORTx | D | Adopting a more convincing gene expression reference from single-cell sequencing | Suitability for some tumor types needs further validation |
MuSiC | D | Adopting a more convincing gene expression reference from single-cell sequencing Available for tissues with intensively correlated cell types | Suitability for some tumor types needs further validation Not available for TPM data as input33 |
FARDEEP | D | A robust machine learning tool eliminating outliers in the dataset Suitable for deconvolution of noisy datasets | Different signature matrix should be adopted according to the type of gene expression data32 |
