Comparing mixed models and Random Forest association tests using naturalGWAS and a Striped Bass SNP dataset

In this study, we used naturalGWAS to test the performance of Zhao's Random Forest method in comparison to an uncorrected classic Random Forest test and two mixed models. We created 100 sets of phenotypes, corresponding to 4 effect sizes and 2, 5, 10, 20, or 30 causal loci, simulated from 7319 empirical SNPs generated from Striped Bass sampled from three distinct locations with high amounts of genetic structure and intraspecific hybrids. Zhao's Random Forest produced the fewest false positives across all tests and performed well when more liberal multiple testing corrections were used, allowing the method to achieve higher power than the next best performing method, CATE. We then used both CATE and Zhao's Random Forest to test for associations between Striped Bass genotype and condition factor, and found no significant loci.