Table 1. Results of cross validation for different weight thresholds (columns 2 to 5).
| Weight Threshold | 1.0 | 0.5 | 0.25 | 0.0 |
|---|---|---|---|---|
| Maximally possible TPs | 3573 | |||
| True positives | 3027 | 3474 | 3461 | 3111 |
| (3372) | (3521) | (3528) | (3485) | |
| False negatives | 546 | 99 | 112 | 462 |
| (201) | (52) | (45) | (88) | |
| False positives | 22 | 37 | 59 | 4631 |
| (18) | (94) | (155) | (7689) | |
| Sensitivity % | 85 | 97 | 97 | 87 |
| (94) | (99) | (99) | (98) | |
| Specificity % | 99 | |||
| F1 % | 91 | 98 | 98 | 55 |
| (97) | (98) | (97) | (47) | |
| Positive predictive value % | 99 | 99 | 98 | 40 |
| (99) | (97) | (96) | (31) | |
A higher threshold enforces utilization of more specific variants but reduces the amount of considered variants. Depending on the threshold (0.0, 0.25, 0.5 1.0), between 3027 and 3474 of the 3573 true relationships between CCLs are successfully recovered. Numbers in brackets show results when the to-be-expected amount of matching variants is set manually to 10 variants; numbers without brackets show statistically estimated background-noise strength (regularized, see methods).