Fig. 2.

Development of immune cell classifier and similarity heatmap. a ROC curve for the immune cell classifier was calculated using the indicated lambda values (shown in different colors and line styles) and 10-fold cross validation. The lambda value that maximized the AUC value was used for subsequent calculations. Elastic-net logistic regression was used to discriminate among ten immune cell types, where the value of the non-zero coefficients (panel b), expression levels (panel c), and similarity map (panel d) for the 452 genes included in the classifier are indicated by color bars for each panel. In panel b, blue to red color scheme indicates coefficients ranging from negative to positive values. Ordering of the genes is the same in panels b and c. In panel c, light blue indicates missing values and the intensity of red color (white/red color scale on the top-left) shows the log base 2 expression level. A color bar on top of this panel was used to separate samples of each cell type. Panel d illustrates the similarity between samples calculated using distance matrix based on same 452 genes. Color bars on the left and bottom sides are to separate samples of each cell type and the top color bar (light blue/pink color scale) shows the intensity of similarity or dissimilarity of samples