TABLE 2.
Performance of single-locus and multilocus methods in mapping two QTL in 100 simulated eight-founder heterogeneous stock populations
| Single locus or multilocus | Threshold
|
False discovery rate (%)e | Chromosome 2 associations (%)f | ||||
|---|---|---|---|---|---|---|---|
| Methoda | Typeb | Units | Valuec | Power (%)d | |||
| Naive single locus | Single | Permutation | Log P | 3.65 (3.43, 3.99) | 99.5 (±0.5) | 92.3 (±0.28) | 52.4 (±0.18) |
| Naive single locus | Single | Parametric bootstrap | Log P | 9.48 (5, 15.5) | 90 (±2.2) | 73.5 (±2.3) | 9.87 (±0.11) |
| Mixed model | Single | Parametric bootstrap | Log P | 3.91 (3.59, 4.33) | 89.5 (±2.3) | 69.5 (±2.1) | 2.41 (±0.056) |
| Single locus | Single | ∼80% power | Log P | 9.88 | 80 (±3.4) | 63.3 (±3.5) | 7.39 (±0.095) |
| Mixed model | Single | ∼80% power | Log P | 5.13 | 80 (±2.9) | 46.6 (±3) | 0.702 (±0.03) |
| Bagging | Multilocus | ∼80% power | Range probability | 0.39 | 80 (±2.8) | 22.4 (±2.6) | 0 (±0) |
| Bagging peaks 1 cM | Multilocus | ∼80% power | Range probability | 0.39 | 80 (±2.8) | 12.8 (±2.5) | 0.0584 (±0.041) |
| Bagging peaks 2 cM | Multilocus | ∼80% power | Range probability | 0.34 | 80 (±2.9) | 23.5 (±2.8) | 0.54 (±0.16) |
| Bagging peaks 4 cM | Multilocus | ∼80% power | Range probability | 0.29 | 80.5 (±2.8) | 30.4 (±2.7) | 1.92 (±0.37) |
| Bagging peaks 8 cM | Multilocus | ∼80% power | Range probability | 0.23 | 80.5 (±2.7) | 43.1 (±2.5) | 7.95 (±0.95) |
| Subagging | Multilocus | ∼80% power | Range probability | 0.4 | 80 (±3) | 16 (±2.8) | 0.00264 (±0.0019) |
| Subagging peaks 1 cM | Multilocus | ∼80% power | Range probability | 0.34 | 80 (±2.8) | 19.8 (±2.8) | 0.204 (±0.077) |
| Subagging peaks 2 cM | Multilocus | ∼80% power | Range probability | 0.27 | 80 (±2.8) | 23.3 (±2.9) | 0.585 (±0.16) |
| Subagging peaks 4 cM | Multilocus | ∼80% power | Range probability | 0.22 | 80.5 (±2.9) | 28.7 (±3) | 2.42 (±0.41) |
| Subagging peaks 8 cM | Multilocus | ∼80% power | Range probability | 0.14 | 80 (±2.8) | 35.9 (±2.9) | 7.33 (±0.91) |
All subagging uses 80% subsamples. The suffix “peaks p cM” denotes where model selection chose among representative peaks from a naive single locus scan rather than among all loci (see simulations).
How the threshold for calculating performance statistics was determined and applied. Permutation: genomewide 5% significance thresholds were calculated separately for each simulation. Parametric bootstrap: thresholds were calculated separately for each simulation. “∼n% power”: scores were pooled for all simulations and the threshold was calibrated to give ∼n% power.
For the permutation and parametric bootstrap results, threshold values are given as “median (lowest, highest).” Otherwise the numbers are the range probability thresholds required to achieve ∼n% power.
The proportion of segments containing QTL that had scores above the threshold, averaged over simulations (SE in parentheses).
The proportion of 6-cM segments with scores above the threshold that did not contain a QTL, averaged over simulations (SE in parentheses).
The proportion of marker intervals on chromosome 2 predicted to contain QTL, averaged over simulations (SE in parentheses).