Skip to main content

Table 1 Performance of drug sensitivity prediction using 500 features selected by ProGENI compared to 500 features selected using baseline schemes, for the LCL dataset

From: Knowledge-guided gene prioritization reveals new insights into the mechanisms of chemoresistance

Treatment ProGENI-SVR > PCC-SVR? (FDR, PIF) ProGENI-SVR > EN-SVR? (FDR, PIF) Average SPCI (ProGENI-SVR) Average SPCI (PCC-SVR) Average SPCI (EN-SVR)
Doxorubicin Yes (3.94E-11, 71.2%) Yes (2.79E-38, 93.2%) 0.660 0.641 0.498
MPA Yes (4.49E-24, 81.2%) Yes (2.63E-12, 71.6%) 0.732 0.714 0.701
TCN Yes (1.30E-19, 78.8%) Yes (2.54E-5, 60.4%) 0.581 0.522 0.539
Everolimus Yes (1.99E-8, 66%) Yes (2.63E-12, 69.6%) 0.595 0.570 0.554
Docetaxel Yes (1.04E-11, 68.4%) Yes (2.84E-7, 66%) 0.608 0.585 0.579
Oxaliplatin No (1.83E-8, 32.4%) Yes (5.44E-12, 68%) 0.676 0.694 0.633
Epirubicin Yes (2.99E-8, 65.6%) Yes (1.61E-12, 66.8%) 0.646 0.628 0.571
NAPQI Yes (1.11E-7, 66.8%) Yes (1.47E-2, 56.4%) 0.597 0.575 0.583
Radiation Yes (4.30E-8, 66%) Yes (2.38E-4, 58.8%) 0.583 0.560 0.561
Carboplatin Yes (2.66E-7, 64%) Yes (6.67E-9, 65.2%) 0.615 0.596 0.578
MTX No (5.52E-3, 42.4%) Yes (2.15E-7, 64%) 0.583 0.588 0.548
Cladribine Yes (5.68E-7, 63.6%) Yes (2.58E-5, 62.4%) 0.622 0.602 0.597
Arac Yes (5.51E-3, 60.45) Yes (9.13E-8, 63.2%) 0.658 0.650 0.629
Paclitaxel Yes (8.94E-7, 62.8%) Yes (6.83E-5, 61.2%) 0.559 0.542 0.538
Arsenic Yes (4.47E-6, 62.4%) Yes (1.47E-2, 57.2%) 0.530 0.513 0.519
6MP No (1.59E-4, 39.2%) Yes (2.58E-5, 62.4%) 0.627 0.638 0.597
Rapamycin Yes (3.56E-7, 61.2%) Yes (8.23E-4, 57.6%) 0.620 0.606 0.602
6TG No (1.62E-19, 24.8%) Yes (5.33E-4, 60.8%) 0.717 0.746 0.694
Metformin Similar (0.514, 48.4%) Yes (1.87E-3, 60.4%) 0.583 0.584 0.564
Fludarabine Similar (0.655, 46.4%) Yes (7.54E-5, 59.2%) 0.539 0.540 0.512
TMZ Similar (2.99E-2, 54%) No (4.23E-3, 41.6%) 0.482 0.471 0.496
Hypoxia Similar (0.406, 45.6%) No (4.41E-2, 42.8%) 0.588 0.590 0.600
CDDP No (2.01E-3, 44%) No (5.30E-4, 39.2%) 0.602 0.607 0.621
Gemcitabine No (1.10E-4, 37.2%) excluded 0.739 0.745 excluded
  1. The FDR is calculated using a two-sided Wilcoxon signed rank test and corrected for multiple tests. The treatments are sorted based on the largest PIF of the improvement obtained using ProGENI compared to any of the baseline schemes. Results in the range 45% < PIF < 55% were considered similar. Gemcitabine was excluded, since for a few training sets the best model trained by EN only included the intercept