Table 1.
Performance of hyperSMURF with different selection strategies for negatives.
| Neg. selection | imb.ratio | n.folds | AUPRC | AUROC |
|---|---|---|---|---|
| Mendelian data | ||||
| ±100 Kb | 1:302 | 116 | 0.7071 | 0.9805 |
| ±500 Kb | 1:1432 | 116 | 0.6279 | 0.9802 |
| ±1000 Kb | 1:2765 | 111 | 0.6161 | 0.9786 |
| TAD | 1:1406 | 125 | 0.6123 | 0.9803 |
| GWAS data | ||||
| ±100 Kb | 1:80 | 1402 | 0.6488 | 0.9840 |
| ±500 Kb | 1:277 | 723 | 0.4796 | 0.9841 |
| ±1000 Kb | 1:409 | 413 | 0.4213 | 0.9851 |
| TAD | 1:269 | 1196 | 0.4792 | 0.9838 |
The first column represents the size of the “genomic window” used to select negatives around each positive or the “TAD-based” negative selection strategy; the second column reports the imbalance between positive and negative examples; the third the number of folds of the “topologically-aware” cross-validation, while the last two columns show the estimated AUPRC and AUROC.