Fig 1.
Distribution of CCLE cell line-drug response values.
A) Violin-plot of existing values for 23 drugs in CCLE; B) Heatmap of cell lines-drug response matrix.
Fig 2.
Distribution of CCLE cell line-drug response values after global effect removal.
A) Violin-plot of 23 drugs’ response distribution in CCLE; B) Heatmap of cell line-drug response matrix.
Fig 3.
Comparison of results with and without GER preprocessing for prediction.
A) Radar plots of PCC_sr and RMSE_sr for 23 drugs in CCLE; B) Box-plots of ave_PCC (ave_PCC_sr), ave_RMSE (ave_RMSE_sr) on the GDSC.
Fig 4.
Effect of the weight of the nonlinear structure α on prediction.
A) ave_PCC and ave_RMSE of network based on drug and cell line latent factors in GDSC; B) ave_PCC and ave_RMSE of network based on drug and cell line latent factors in CCLE.
Table 1.
Comparison of ave_PCC (ave_PCC_sr), ave_RMSE (ave_RMSE_sr) of different methods for 10-fold cross-validation on CCLE dataset.
Table 2.
Comparison of ave_PCC (ave_PCC_sr), ave_RMSE (ave_RMSE_sr) of different methods for 10-fold cross-validation on GDSC dataset.
Fig 5.
Scatter-plot between the drug responses predicted by the DBDNMF model and the original responses in CCLE.
Table 3.
The ave_PCC (ave_PCC_sr), ave_RMSE (ave_RMSE_sr) of NeuMF vs. DBDNMF for cold-start scenarios on CCLE dataset.
Table 4.
The ave_PCC (ave_PCC_sr), ave_RMSE (ave_RMSE_sr) of TGSA, NeuMF vs. DBDNMF for cold-start scenarios on GDSC dataset.
Table 5.
Results of drug response prediction on the TCGA dataset.
Fig 6.
The comparison of drug response prediction.
A) The Receiver Operating Characteristic Curve; B) The Precision-Recall Curve.
Fig 7.
Comparison histogram of PCC_sr and RMSE_sr for the drug targets that are involved in the PI3K pathway of GDSC.
A) Comparison histogram of PCC_sr for drugs targeting the PI3K pathway; B) Comparison histogram of RMSE_sr for drugs targeting the PI3K pathway.
Fig 8.
Association of drug sensitivity with cancer gene mutations is consistent for predicted and existing responses.
A) RB1 mutated cell lines are more sensitive to the drug IOX2; B) RB1 mutated cell lines are more resistant to the drug Dactolisib; C) BARF mutated cell lines are more sensitive to the drug Refametinib.
Fig 9.
Hierarchical clustering of ALL cancer cell lines to 11 drugs come from the cluster 2.
Fig 10.
Hierarchical clustering of MM cancer cell lines to 11 drugs come from the cluster 2.
Fig 11.
Network structure. The input layer inputs a low-dimensional unknown potential matrix.
The hidden layers are divided into linear hidden layers and nonlinear hidden layers. The output layer outputs the completed cell line-drug response matrix.