Table 3.
Summary of drug response prediction methods
| Method | Prediction problem | Model | Parameter selection / Inference | Model evaluation | Performance evaluation | Note |
|---|---|---|---|---|---|---|
| SRMF [21] | Drug response prediction, drug repositioning | MF | Alternating minimization | 10-fold CV | PCC, RMSE | Drug similarity information does not contribute to prediction performance. |
| HMF [26] | Drug response prediction, prediction of cancer driver genes | BMF | Gibbs sampling | 10-fold CV | MSE | Automatic relevance determination eliminates the need to perform model selection. |
| CaDRReS [28] | Drug response prediction, drug and cell line clustering, drug-pathway associations | MF | Gradient descent | 5-fold CV | SCC, NDCG | NDCG metric is preferred for ranking drugs, which is claimed useful from a clinical perspective. |
| QSAR [29] | Drug response prediction | KBMF | Variational inference | Nested 8-fold CV | MSE, R2, PCC | All information yields more powerful performance than drug descriptors or targets alone. |
| cwKBMF [25] | Drug response prediction, drug-pathway associations | KBMF | Variational inference | Nested 5-fold CV | SCC | The model recognizes component specific relationships between multiple sources and drug responses. |
| pairwiseMKL [22] | Drug response prediction, drug-target prediction | KRR | Conjugate gradient | Nested 10-fold CV | RMSE, PCC, F1 score | The model known as the first method for time and memory-efficient learning with multiple pairwise kernels. |
| Dual-Layer [30] | Drug response prediction | NBR | Sum of squared errors | LOOCV | PCC, RMSE, NRMSE | Drug similarities are more informative than cell line similarities. However, using both gives better results overall. |
| Stanfield’s method [20] | Drug response prediction | NBC | Random walk | LOOCV | AUC | Known as the first drug response prediction method directly integrates response data, mutation data and PPIs into a network. |
| HNMDRP [31] | Drug response prediction | NBC | Information flow-based algorithm | LOOCV | AUC | PPI and gene–gene correlation information have more vital role than other sources. |
Note: MF = Matrix factorization; BMF = Bayesian matrix factorization; KBMF = Kernel Bayesian matrix factorization; KRR = Kernel ridge regression; NBR = Network based regression; NBC = Network based classification; CV = Cross validation; LOOCV = Leave-one-out cross validation; PCC = Pearson correlation coefficient; RMSE = Root mean square error; MSE = Mean square error; SCC = Spearman correlation coefficient; NDCG = Normalized discounted cumulative gain; R2 = Coefficient of determination; NRMSE = Normalized root mean squared error; AUC = Area under curve; PPI = Protein–protein interaction.