Fig. 8: Drug response with tumor heterogeneity.
From: Deep generative neural network for accurate drug response imputation
A Distribution of gene t-scores for all drugs, with stroma genes, highlighted. Stroma genes were found to be negatively associated with several EGFR inhibitors, such as AZD6244, lapatinib, and PD-0325901. B Distribution of gene t-scores with Tumor Inflammation Score (TIS) signatures. C Investigation of the association between BRAF mutations and PLX4720 response in SKCM. SKCM samples with BRAF V600 mutations (labeled as “Mut” on the left panel, nMut = 157) did not show a significantly high response to PLX4720 compared to other SKCM samples that had no mutations in BRAF, HRAS, KRAS, NRAS, or EGFR (labeled as “WT” on the left panel, nWT = 59). However, this was explained using the “downregulation of ERBB2 ERBB3 signaling” pathway activity. In the subgroup where this pathway showed low activity (the lower quartile of all samples ordered by increasing pathway activity, labeled Q25, nQ25 = 91, including nMut = 48 for samples with BRAF V600 and nWT = 43 for samples without BRAF V600), samples with BRAF mutations had a higher response, implying that the pathway had a confounding effect on the response. In the middle subgroup (Q25_75), there are nQ25_75 = 180 samples with nMut = 80 samples harboring BRAF V600 mutations and nWT = 100 samples harboring not. In the higher quantile subgroup (Q75), there are nQ75 = 91, including nMut = 29 for samples with BRAF V600 and nWT = 62. A two-sided t test was conducted in each subgroup to compare the predicted response between the samples with the BRAF V600 mutation (nMut) and samples without BRAF V600 mutations (nWT). D Investigation of the association between BRAF mutations and PLX4720 response in THCA. THCA samples were grouped into three subtypes according to the original publication: Braf-like (n = 271, where n = 224 harboring BRAF V600 mutations and n = 47 not), Ras-like (n = 67, where all having not the BRAF V600 mutation), and others (n = 89, where 63 harboring BRAF V600 mutations and n = 26 not). When stratifying the samples by the EGFR downregulation pathway activity, the samples in each subgroup were as below: nMut = 42 and nWT = 81 in the Q25 subgroup, nMut = 170 and nWT = 74 in the Q25_75 subgroup, and nMut = 75 and nWT = 48 in the Q75 subgroup. In each subgroup, a two-sided t test was conducted to compare the response between the samples with BRAF V600 mutations (nMut) and samples without (nWT). In the boxplots (C, D), each box shows the interquartile range (IQR between Q1 and Q3) for the corresponding set. The central mark (horizontal line) shows the median.
