Fig. 3: LUAD classification by RAS84.
From: RAS oncogenic activity predicts response to chemotherapy and outcome in lung adenocarcinoma
a Heatmap showing clustered RAS84 genes and TCGA LUAD cohort patients. Patients are shown as rows, genes as columns. Patients have been clustered into five RAS activity groups (RAGs) by hierarchical clustering using a ward.D2 agglomeration method. Aggregate RAS84-Index (RI) scores are shown to the right of the main heatmap. Genome variants with a significant non-random distribution across the RAGs are shown in the nine columns on the right (chi-square fdr < 0.05). These mutations are used to characterise the five clusters shown by the labels on the right (P, TP53; KL, KRAS/LKB1(STK11); K, KRAS; KP, KRAS/TP53; KC, KRAS/CDKN2A). KRAS mutants are shown in dark red. Parent signature membership is shown in grey at the bottom of the heatmap. b The percentage of KRAS mutations per RAG broken down by specific KRAS mutation type. c Log-likelihood values from a GLM fit (family = binomial) of KRAS mutation status across the five RAGs. d Bar plots showing the percentage of patients per RAG with EGFR, TP53, STK11 mutations, CDKN2A deletion, KEAP1, RB1, ATM and CTNNB1 mutations found to be significantly associated with any one RAG (fdr < 0.05). e EGFR, KRAS and TP53 mutation percentages found to be significantly associated with any one RAG from the Seoul cohort. f Boxplots showing The Cancer Protein Atlas (TCPA) RPPA MEK1 and ERK1/2 phosphorylation level distributions across RAGs. Significance levels are shown compared to RAG-0 derived by linear model fit (n = 349; RAG-0 n = 57, RAG-1 n = 79, RAG-2 n = 61, RAG-3 n = 91, RAG-4 n = 61; ****P ≤ 0.0001, ***P ≤ 0.001, **P ≤ 0.01, n.s.= P > 0.05) The box shows the median and IQR, the whiskers indicate ± 1.5 x IQR, outliers lie outside this range. g Heatmap showing variant mean RAS84 gene expression clusters across the five RAGs.
