Fig. 3
Associations between factors with tissue type, pathways and drug responses. a Predictive performances of ENs trained with raw features (x-axis) and factors (y-axis) for 265 compounds in Pearson correlation. Usage of mutation data and the total number of features selected in the raw feature-based are indicated by node color and size. b The relative prediction performance of ENs trained with raw features (raw-EN) and factors (factor-EN; x-axis), compared with the reconstruction efficiency of the features in raw-EN model (average explained variation of the selected features by WON-PARAFAC; y-axis) in a scatter plot. Node color/size indicates the use of mutation data by raw-EN and performance of factor-EN, respectively. c–e EN coefficient (top; y-axis) and relative contribution (bottom; y-axis) of the factors (x-axis) predicts response to Afatinib (c), Camptothecin (d) and SN-38 (Irinotecan; e). The dark bar indicates the six factors used in the downstream analysis. For Camptothecin and SN-38, we selected the union of the top five factors with the largest absolute EN coefficients from in the two EN models. f Comparison of response to Camptothecin (left panel) and SN-38 (Irinotecan; right panel) between cell lines with higher loadings on sensitivity (predicted sensitive; left boxplot) and resistance factors (predicted insensitive; right boxplot), respectively. Standard notations are used for elements of the boxplot (i.e. upper/lower hinges: 75th/25th percentiles; inner-segment: median; and upper/lower whiskers: extension of the hinges to the largest/smallest value at most 1.5 times of interquartile range)
