Table 5. Evaluation of the accuracy of Crinet- and Hermes-inferred ceRNA interactions based on the shRNA-mediated gene knockdown experiment.
| 96h Timepoint | 144h Timepoint | Overall | |
|---|---|---|---|
| Crinet | 48/77 ≈ 62% | 44/77 ≈ 57% | 92/154 ≈ 60% |
| Hermes 1.run | 412/897 ≈ 46% | 441/913 ≈ 48% | 853/1810 ≈ 47% |
| Hermes 2.run | 199/466 ≈ 43% | 217/481 ≈ 45% | 416/947 ≈ 44% |
| Hermes 3.run | 109/269 ≈ 41% | 121/273 ≈ 44% | 230/542 ≈ 42% |
| Hermes 4.run | 65/143 ≈ 45% | 68/144 ≈ 47% | 133/287 ≈ 46% |
| Hermes 5.run | 33/69 ≈ 48% | 39/71 ≈ 55% | 72/140 ≈ 51% |
Analysis to check the accuracy of inferred ceRNA interactions using LINCS-L1000 shRNA-mediated gene knockdown experiment in breast cancer cell line. Based on the ratios of gene expression fold-change following the knockdown of its ceRNA partners to following the genes that are not its ceRNA partners for each perturbagen ceRNA, the accuracy of a ceRNA network was accepted as the percentage of ceRNAs whose ratios were smaller than 1 with respect to all ceRNAs. We calculated the accuracy separately for each different timepoint (96h & 144h) and combined timepoints as the overall. Hermes’s x.run had 10−(x+1) for the significance of the common miRNA size and 10−(x+2) for the significance of conditional regulation.