Plink And Missing Values

Question

[RawRaw]

Do we need to impute the missing values before finding the significant markers in genome wide association studies? Specifically, using PLINKs permutation tests to find the adjusted p-values that state the significant markers. If not, how does PLINK handles the missing values?

Answer 1

[Fabio Marroni]

It's not mandatory to impute missing data. I guess PLINK will treat your missing genotypes as missing data, both in the association analysis and in the permutations. Obviously, having missing data may decrease your power. Also, having a lot of missing data in a sample (or in one SNP) may be indicative of low quality.

[RawRaw]

" I guess PLINK will treat your missing genotypes as missing data, both in the association analysis and in the permutations." How, and why not imputing them?

[chrchang523]

Nobody is stopping you from imputing with other software. But the chi-square/Fisher's exact tests used by PLINK's basic association analysis commands work just fine with (not-terribly-biased) missing data.

Note that PLINK 1.9's --assoc and --model permutation tests are hundreds to thousands of times faster than PLINK 1.07's.

[RawRaw]

@chrchang523 So, PLINK just neglects the missing SNPs? This means the numerical perturbation caused be the missing SNPs is negligible and could be discarded in MaxT permutation tests (assuming that there are very few missing SNPs, and not-terribly-biased).

[RawRaw]

Since PLINK 1.9's permutation procedures extend work by Brian Browning (PRESTO) and Roman Pahl (PERMORY), it would be much, much faster than 1.7s.