Fig. 6: Drug similarity.
From: Deep generative neural network for accurate drug response imputation
A Hierarchy clustering results of the 24 CCLE drugs using four sets of samples: the observed drug response (set 1), the predicted drug response in cell lines with observed data (set 2), the predicted drug response in all 1100 cell lines (set 3), and the predicted drug response in all TCGA cancer samples. B Heatmap showing the association patterns between each drug and the tumor mutation burden (TMB) in CCLE. In each cell, red indicates a positive association (i.e., responders of the cancer type tended to have a high TMB), while blue indicates a negative association (i.e., responder tended to have a low TMB). For each cell, a two-sided t test was conducted by comparing the log10(TMB) of samples from the group with high sensitivity (i.e., “responders”, defined as the top 25% samples ordered by decreasing predicted response to the corresponding drug) and the log10(TMB) of the remaining 75% samples. The color was determined by whether the sensitive group had a higher average TMB than the other group. C, D Demonstration of TMB distribution in responders (the 25% samples with the highest response) and non-responders (the remaining 75% samples) using two cytotoxic drugs: irinotecan (C) and topotecan (D). Each box shows the interquartile range (IQR between Q1 and Q3) for the corresponding set. The central mark (horizontal line) shows the median and the dots show the rest of the distribution based on IQR (Q1−1.5 × IQR, Q3 + 1.5 × IQR). E Distribution of the association between GDSC drugs and TMB. Because TMB varied dramatically across cancer types, we conducted the test within each cancer type. Each dot represents a GDSC drug in a cancer type. The P value was similarly calculated as in (B). x axis is the log2 form of fold change (FC) defined as the average TMB of the responder group over the average TMB of the nonresponder group. Y axis indicates –log10 form of the unadjusted P value (for plotting only). Red and blue dots indicate significant associations (PBH < 0.05 and log2FC > 0.2). F, G Enrichment of drug classes in the group of drug-cancer type associations with negative association patterns (blue dots in E) and with positive association patterns (red dots in E). The P values were obtained from a two-sided Fisher’s exact test to assess whether a drug class was overrepresented with the negative/positive associations.
