Table 5.
Classification performance for different cell lines
| Cell Type | AUROC | AUPRC | ||
|---|---|---|---|---|
| DeepEnhancer | gkmSVM | DeepEnhancer | gkmSVM | |
| GM12878 | 0.874 | 0.784 | 0.875 | 0.819 |
| H1-hESC | 0.923 | 0.869 | 0.919 | 0.861 |
| HepG2 | 0.882 | 0.800 | 0.883 | 0.827 |
| HMEC | 0.903 | 0.848 | 0.907 | 0.892 |
| HSMM | 0.904 | 0.830 | 0.910 | 0.856 |
| HUVEC | 0.898 | 0.824 | 0.905 | 0.870 |
| K562 | 0.883 | 0.794 | 0.886 | 0.799 |
| NHEK | 0.888 | 0.809 | 0.893 | 0.840 |
| NHLF | 0.909 | 0.848 | 0.910 | 0.869 |
| p-value | 1.9e-3 | 1.9e-3 | ||
We compare the performance of our DeepEnhancer model and gkmSVM on 9 cell types using two measures: area under receiver operating characteristic curve (AUROC) and area under precision-recall curve (AUPRC). The last row shows the p-value result of the binomial exact test, which makes us choose the alternative hypothesis that DeepEnhancer has a larger AUC score than gkmSVM