Differences In Significant Snps Between Plink And Pasw Using Logistic Regression

Question

[Philipp Bayer]

Hello everyone,

for fun I checked whether there are any differences between PLINK and PASW/SPSS when doing a logistic regression (and standard parameters) on a binary trait. Using the same dataset, I get quite different significant SNPs - they sometimes overlap slightly, but sometimes the most significant SNPs in PLINK don't even appear in the model that PASW builds.

For testing I don't use any SNP-specific cleaning-steps etc., just wanted to check how both programs compare.

Does anybody have any explanation?

Thanks!

Answer 1

[ff.cc.cc]

Probably, PLINK and SPSS have two different default model approach. From the plink online help:

For the additive effects of SNPs, the direction of the regression coefficient represents the effect of each extra minor allele

Plink usually works on SNPs as a continous predictor, maybe SPSS manage your SNPs as a factor.

[Philipp Bayer]

So in doing a GWAS, I should "trust" Plink more than SPSS?

[ff.cc.cc]

You could trust the plink logistic reg. approach if you believe in additive effects of the rare allele (more alleles -> more effect in almost linear way). If you believe there could be other possible explanation for your outcome (e.g. recessive, dominant, co-dominant...) you could treat SNPs as factors (each allele has its own effect on the outcome).

p.s. if you do not worry about covariates like age sex etc... but only about genetic component consider the --model option of plink