Fig. 1

Overview of computational pipelines. a ProGENI: an RWR is used to obtain a vector representation of each gene and is used to perform a network transformation on the gene expression. The response-correlated genes (RCGs) are identified as 100 genes whose transformed expressions have the highest absolute PCC with the drug response. An RWR is used to score each gene based on similarity to the RCG. These scores are then normalized to remove the network bias. b Robust ranking: 80% of the cell lines are selected randomly and used with a prioritization method to obtain a ranked list of genes. This procedure is repeated r times and the acquired ranked lists are aggregated to obtain a final ranked list. c Cross-validation scheme: a nonlinear support vector machine is trained on the training set using the top 500 genes to predict drug sensitivity of the test set and evaluate the accuracy of prediction