Fig. 3

ANN-SCGP accurately predicted SF2. A Illustration of occlusion test. The genes with top 25% occlusion scores had significantly higher SF2-predictive C-indexes than others in CCLE B and GDSC C. Gene ontology analysis was performed by top 25% occlusion score genes in CCLE D and GDSC E. We clustered samples into two groups (C1 & C2) via K-means of genes with top 25% occlusion scores in CCLE F and GDSC G. C2 samples had high SF2 in CCLE H and GDSC I. C1 was linked to high signals of double-strand break repair in CCLE J and GDSC K. L The horizontal coordinate was the enumeration of all combinations of node-based and gene-based sample similarity from different groups (controls). The vertical coordinate was the Pearson's correlation of node-based and gene-based sample similarity from controls minus that from same group. RMSE, root mean squared error; CCLE, Cancer Cell Line Encyclopedia; GDSC, Drug Sensitivity in Cancer; LM, linear regression model; SVM, support vector machine; RF, random forest; RSI, radiosensitivity index