Table 3.
Accuracy of GWIS Based on GWAS of Dichotomous Traits
| R-squared | SNP effect | Latent GWAS | Dichotomous GWAS | Ratio 1 | Latent GWIS | Dichotomous GWIS | Ratio 2 |
|---|---|---|---|---|---|---|---|
| 0.0001 | 0.0050 | 0.0048 | 0.0089 | 1.8383 | 0.0059 | 0.0101 | 1.7147 |
| 0.0001 | 0.0100 | 0.0109 | 0.0167 | 1.5263 | 0.0125 | 0.0179 | 1.4290 |
| 0.0002 | 0.0150 | 0.0167 | 0.0279 | 1.6740 | 0.0179 | 0.0300 | 1.6766 |
| 0.0002 | 0.0200 | 0.0198 | 0.0346 | 1.7509 | 0.0234 | 0.0421 | 1.8034 |
| 0.0003 | 0.0250 | 0.0241 | 0.0430 | 1.7855 | 0.0281 | 0.0514 | 1.8283 |
| 0.0004 | 0.0300 | 0.0285 | 0.0511 | 1.7951 | 0.0324 | 0.0590 | 1.8210 |
| 0.0006 | 0.0350 | 0.0350 | 0.0638 | 1.8226 | 0.0410 | 0.0741 | 1.8089 |
| 0.0007 | 0.0400 | 0.0393 | 0.0706 | 1.7971 | 0.0460 | 0.0834 | 1.8122 |
| 0.0009 | 0.0450 | 0.0452 | 0.0814 | 1.8027 | 0.0514 | 0.0933 | 1.8152 |
| 0.0010 | 0.0500 | 0.0494 | 0.0881 | 1.7848 | 0.0562 | 0.1002 | 1.7830 |
| 0.0013 | 0.0550 | 0.0559 | 0.1013 | 1.8104 | 0.0646 | 0.1178 | 1.8237 |
| 0.0014 | 0.0600 | 0.0588 | 0.1058 | 1.7995 | 0.0682 | 0.1224 | 1.7950 |
| 0.0017 | 0.0650 | 0.0660 | 0.1193 | 1.8073 | 0.0751 | 0.1367 | 1.8197 |
| 0.0019 | 0.0700 | 0.0699 | 0.1249 | 1.7853 | 0.0806 | 0.1443 | 1.7914 |
| 0.0023 | 0.0750 | 0.0760 | 0.1337 | 1.7594 | 0.0862 | 0.1529 | 1.7742 |
| 0.0025 | 0.0800 | 0.0797 | 0.1402 | 1.7597 | 0.0915 | 0.1609 | 1.7591 |
| 0.0028 | 0.0850 | 0.0851 | 0.1533 | 1.8022 | 0.0970 | 0.1743 | 1.7969 |
| 0.0032 | 0.0900 | 0.0907 | 0.1633 | 1.8000 | 0.1028 | 0.1844 | 1.7935 |
| 0.0035 | 0.0950 | 0.0949 | 0.1691 | 1.7812 | 0.1093 | 0.1935 | 1.7709 |
| 0.0038 | 0.1000 | 0.0986 | 0.1763 | 1.7884 | 0.1137 | 0.2032 | 1.7874 |
| 0.0085 | 0.1500 | 0.1495 | 0.2671 | 1.7861 | 0.1707 | 0.3039 | 1.7806 |
| 0.0149 | 0.2000 | 0.1969 | 0.3506 | 1.7804 | 0.2259 | 0.4018 | 1.7790 |
| 0.0232 | 0.2500 | 0.2453 | 0.4375 | 1.7833 | 0.2829 | 0.5030 | 1.7777 |
| 0.0325 | 0.3000 | 0.2890 | 0.5188 | 1.7950 | 0.3356 | 0.6014 | 1.7920 |
| 0.0442 | 0.3500 | 0.3356 | 0.6060 | 1.8059 | 0.3914 | 0.7064 | 1.8046 |
| 0.0565 | 0.4000 | 0.3765 | 0.6917 | 1.8370 | 0.4416 | 0.8115 | 1.8377 |
| 0.0707 | 0.4500 | 0.4184 | 0.7725 | 1.8464 | 0.4936 | 0.9118 | 1.8474 |
| 0.0862 | 0.5000 | 0.4581 | 0.8589 | 1.8749 | 0.5431 | 1.0204 | 1.8787 |
Results obtained from simulated GWAS on a dichotomous outcome and simulated GWIS performed on the continuous latent variable produces effect sizes that differ approximately by a constant multiple, relative to the dichotomous GWAS and dichotomous GWIS. The reported values here are means over 500 runs, each containing N = 10,000 individuals. Ratio 1 reflects the ratio of the mean effect size from the dichotomous GWAS divided by the mean effect size of the latent GWAS, and ratio 2 reflects the mean effect of the dichotomous GWIS divided by the mean effect of the latent GWIS. Given small effects, the results are approximately equal up to a multiplicative constant. For very large SNP effect sizes, much larger than is usual for polygenic traits, these ratios no longer appear to be constant.