[R] Fastest way to do HWE.exact test on 100K SNP data?

[dhinds at sonic.net]

Anna Pluzhnikov <apluzhni at bsd.uchicago.edu> wrote:
> Hi everyone,


> I'm using the function 'HWE.exact' of 'genetics' package to compute
> p-values of the HWE test. My data set consists of ~600 subjects
> (cases and controls) typed at ~ 10K SNP markers; the test is applied
> separately to cases and controls. The genotypes are stored in a list
> of 'genotype' objects, all.geno, and p-values are calculated inside
> the loop over all SNP markers.

Just to concur with the previous two posters: when I've needed to
calculate lots of HWE values, I've done one of two things:

1.  Use a faster test: either the chisq test or a likelihood ratio
    test.  Optionally, you could use the exact test when one of the
    allele counts is very small, and use an asymtotic test in other
    cases.  I often just use the fast test on everything.  Especially
    since you are doing a permutation analysis on the test values, the
    exact test may not be buying you anything.

2.  Caching the HWE values works great if you can get some reuse of
    previously calculated values.  If you're calculating for ~300
    cases and ~300 controls, there would be 600*601/4 or only ~90K
    possible sets of AA/AB/BB allele counts assuming complete data
    (you can flip the counts around so that the "AA" count is always
    for the more frequent allele); with missing data there can be many
    more, but most of those possibilities will never be observed under
    the null hypothesis of HWE.  And since you are computing roughly
    10 million HWE values, you'll have a lot of reuse of previously
    calculated values.

-- Dave