SF-Relate achieves near-perfect accuracy for identifying close relatives in the UK Biobank and All of Us data sets
| Data set | Recall (%, counts) | Precision (%, counts) | % of comparisons w.r.t. all-pairwise | ||||
|---|---|---|---|---|---|---|---|
| Relatedness degree | Overall | ||||||
| Zero | First | Second | Third | ||||
| UKB-200K | 100.0% | 100.0% | 99.8% | 94.9% | 97.0% | 98.5% | 0.13% |
| 16/16 | 4702/4702 | 1709/1711 | 8475/8925 | 14,902/15,354 | 14,902/15,129 | ||
| UKB-100K | 100.0% | 100.0% | 100.0% | 95.1% | 97.2% | 98.7% | 0.26% |
| 6/6 | 1243/1243 | 404/404 | 2169/2279 | 3822/3932 | 3822/3872 | ||
| AoU-20K | 100.0% | 100.0% | 100.0% | 94.1% | 98.0% | 100.0% | 1.28% |
| 14/14 | 209/209 | 93/93 | 145/154 | 461/470 | 461/461 | ||
[i] Ground-truth relatedness degrees for recall and precision metrics are obtained using the KING method and assigning each sample to the lowest degree of relatedness observed. SF-Relate obtains accurate results while performing only a small fraction of comparisons compared with all-pairwise comparison between data sets. (w.r.t.) With respect to.