TABLE 4.
Performance of LogitDA and KNNDA compared to the other methods in terms of prediction AUC across four targeted therapies and six chemotherapies.
| Method | Geeleher et al. (2014) | MOLI complete (expression data) | MOLI complete (multi omics data) | LogitDA | KNNDA |
|---|---|---|---|---|---|
| Drug (test dataset) | |||||
| Docetaxel (GSE6434) | 0.74 a | 0.31 a | X | 0.76 ± 0.019 | 0.87 b ± 0.010 |
| Erlotinib (GSE33072) | 0.60 | 0.73 | X | 0.94 ± 0.004 | 0.90 ± 0.004 |
| Sorafenib (GSE33072) | 0.45 | 0.65 | X | 0.70 ± 0.003 | 0.71 ± 0.044 |
| Cetuximab (PDX) | 0.58 | 0.51 | 0.53 | 0.93 ± 0.006 | 0.95 ± 0.018 |
| Erlotinib (PDX) | 0.67 | 0.39 | 0.63 | 1.00 ± 0.000 | 1.00 ± 0.000 |
| Gemcitabine (PDX) | 0.59 | 0.52 | 0.64 | 0.83 ± 0.015 | 0.62 ± 0.006 |
| Paclitaxel (PDX) | 0.52 | 0.69 | 0.74 | 0.68 ± 0.022 | 0.65 ± 0.073 |
| Cisplatin (TCGA) | 0.62 | 0.75 | 0.66 | 0.62 ± 0.012 | 0.67 ± 0.028 |
| Docetaxel (TCGA) | 0.59 | 0.63 | 0.58 | 0.81 ± 0.005 | 0.77 ± 0.041 |
| Gemcitabine (TCGA) | 0.53 | 0.64 | 0.65 | 0.62 ± 0.004 | 0.68 ± 0.031 |
a
The initial input genes of the work of Geeleher et al. (2014) and MOLI complete were the same as those of LogitDA and KNNDA, as we could not assess the input genes of the work of Geeleher et al. (2014). The parameters of MOLI complete were optimized using the training data.
b
The bold-faced values indicate the highest prediction AUC among the five methods for a drug.