Table 3.
Runtime estimates of the six methods
| Step | Sample | BOLT-REML | GCTA-GREML | RHE | SUM-RHE | LDSC | SumHer |
|---|---|---|---|---|---|---|---|
| Reference statistic estimation | 281k | — | — | — | 14,286.4 | 3120.8 | 697.3 |
| GWAS summary | 10k | — | — | — | 7.0 | 7.0 | 7.0 |
| Heritability estimation | 10k | 589.6 | 847.3 | 590.6 | 1.6 | 4.0 | 2.9 |
| Reference statistic estimation | 241k | — | — | — | 14,686.0 | 2699.9 | 1180.7 |
| GWAS summary | 50k | — | — | — | 17.9 | 17.9 | 17.9 |
| Heritability estimation | 50k | — | — | 3032.6 | 1.3 | 4.0 | 1.6 |
-
We ran each method on 10 replicates to measure wall clock time. For methods or tools that allow multithreading (BOLT/GCTA-GREML, SumHer, PLINK 2.0), we used six threads, run on the UCLA Hoffman2 computing nodes. SUM-RHE trace summaries were estimated by running the original RHE-mc codes (which does not support multithreading). PLINK 2.0 was used for calculating the GWAS summary statistics. All measurements are in seconds.
