Skip to main content
Fig. 4 | BMC Systems Biology

Fig. 4

From: Reconstructing cancer drug response networks using multitask learning

Fig. 4

P-values for survival models fit using mRNA expression of genes in four sets: genes identified by the multi-task learning method for each drug, COSMIC cancer genes, all genes present in mRNA expression data, and single-task genes. For COSMIC, all genes, and single-task genes, 100 random subsets of available genes are chosen; each random subset contains the same number of genes as the multi-task set for a specific drug. Models are fit to a random training set chosen from 80% of patients, risk scores are calculated for training set and validation set samples, and the median risk in the training set is used as a threshold to divide validation set samples into two groups. P-values are computed from the difference in survival between the two groups of validation set samples. a shows results for paclitaxel, b shows docetaxel, c shows doxorubicin

Back to article page