I'm having some trouble running regenie (v2.2.4) on a quantitative phenotype for a large cohort. I'm testing a standard height GWAS with heights rounded to the nearest integer. I've tried a few different tests to see where the issue lies.
Converting the phenotype to binary and modifying regenie options to match returns reasonable results, as does running the quantitative cohort with plink.
I've tried running without the covariate file, using older regenie versions (v2.0.2 and v2.2), and some experiments with flag options, all to no avail. The median log10 P-value returned on these regenie runs is around 700, with a maximum of 90000, far too high. (in plink, median~0.3, max~90)
Options used for the no covariate run, filenames/paths removed: --step 1 \ --bed (...) \ --phenoFile (...) \ --keep (...) \ --extract (...) --bsize 1000 \ --threads 8 \ --out (...)
--step2 \ --bgen (...) \ --sample (...) \ --phenoFile (...) \ --pred (...) \ --keep (...) \ --gz \ -pThresh 0.01 \ -bsize 1000 \ --threads 8 \ --verbose \ --out
Any ideas on what might be going wrong?