Logistical regression on binary SNP dataset

Question

[gammyknee]

Hi All,

I have a question regarding the statistical test that I need to perform on my dataset (over 1,000 SNPs and about 300 sequences). Its primarily made up of SNP data (as binary 0/1 - as in SNP being found or absent) and I what to find the combination of SNPs that give the highest activity. Ive had a look at logistical regression functions (which I read are perfect for binary data of this calibre), but most examples Ive seen deal with the total number of "0" or "1" observations in the dataset, making it more of an additive approach (i.e. the more SNPs you have, leads to the highest activity). But what I'm after is identifying the particular SNPs or combination of SNPs that produce the greatest effect on the activity (may or may not be a large amount of SNPs that give the highest activity).

Here's an example of the dataset.

Sequence       SNP1         SNP2          SNP3        ....           Activity
1                      0                  0                    1                           760
2                      1                  0                    0                           123
3                      1                  1                    0                           1009
4                      1                  0                    1                           6
.....

Any help or advice would be much appreciated. Im learning R, so any examples in that language would be really helpful

Answer 1

[Philipp Bayer]

Hi,

The package CGEN includes the method snp.logistic to run a logistic regression on a single SNP, so you'd have to iterate over all of your SNPs in a loop. The method snp.match.logistic looks like it can run on a set of SNPs.

I personally prefer PLINK since PLINK can also correct for multiple testing when running regressions and automatically splits into case/control, but then you'd first have to change your data into PLINK's format which is a bit annoying.