Table 2.
Performance comparisons on recall, precision, and F-measure (100 SNPs, 1600 sample size). Runtimes are displayed as a mean value of models (multiplicative models: DME-1–DME-4, threshold models: DME-5–DME-8, and concrete models: DME-9–DME-12).
| Models | Algorithms | Recall | Precision | F-Measure | Runtime(s) |
|---|---|---|---|---|---|
| Multiplicative model | FDHE-IW | 36.5% | 40.2% | 35.7% | 2.95 |
| MACOED | 68.5% | 90.8% | 71.7% | 10.8 | |
| BEAM | 14.8% | 10.3% | 7.7% | 8.52 | |
| BOOST | 0.5% | 62.5% | 0.5% | 0.6 | |
| SNPHarvester | 0.3% | 50.0% | 0.5% | 2.97 | |
| Threshold model | FDHE-IW | 86.9% | 66.4% | 71.0% | 2.95 |
| MACOED | 98.0% | 83.0% | 89.3% | 11.40 | |
| BEAM | 84.5% | 68.3% | 60.5% | 8.52 | |
| BOOST | 34.8% | 99.0% | 37.1% | 0.95 | |
| SNPHarvester | 51.8% | 47.0% | 29.0% | 2.92 | |
| Concrete model | FDHE-IW | 96.3% | 91.0% | 93.3% | 2.95 |
| MACOED | 98.8% | 84.8% | 91.0% | 11.59 | |
| BEAM | 81.3% | 62.3% | 70.1% | 8.52 | |
| BOOST | 66.3% | 87.3% | 69.7% | 0.66 | |
| SNPHarvester | 91.3% | 57.3% | 67.7% | 2.90 |
FDHE-IW: A fast approach for detecting high-order epistasis with interaction weight; MACOED: Multi-objective ant colony optimization epistasis detection; BEAM: Bayesian epistasis association mapping; BOOST: Boolean operation-based screening and testing.