Reproducibility and concordance with alternative cell type quantification methods

Cell typeCorrelation with IHCRoot Mean Squared Error from IHCCorrelation with flowRoot Mean Squared Error from FlowMean pairwise similarity statistic in TCGASD due to technical noise (log2 scale)Proportion of variance due to noise
Cytotoxic cells0.690.08130.001
Exhausted CD80.440.16240.0062
Mast cells0.740.19490.0086
NK CD56dim cells0.470.0710.400.23470.1073
NK cells0.510.1180.470.19380.017
Th1 cellscNA0.22120.0304
CD8 T cells0.531.50.780.1380.510.18420.0045
CD4 cellsb0.65b0.752

aUsed to normalize the other cell types; 0.78 and 0.064 are the highest correlation and lowest RMSE observed between gene expression and flow for any T-cells vs. other cell type contrast

bCalculated as the T-cell score minus the CD8 cell score

cOnly one marker gene; quality impossible to assess in expression data alone

Root mean squared errors are calculated from log2-scale abundance measurements. The mean pairwise similarity statistic measures how well a gene set’s co-expression pattern adheres to the co-expression pattern of ideal marker genes, with a value of 1 indicating perfect correlation with a slope of 1. The standard deviation (SD) and proportion of variance due to noise were calculated from triplicate gene expression assays from tumor sample RNA