Snps Selection/Filtering

Question

[TitoPullo]

I have a dataset with 2 Millions SNPs associated to 1000 individuals. Each SNP is represented, for each individual, as: 0,1 or 2 (the number of minor alleles). I'd like to reduce the number of SNPs in order to use them as attributes with an SVM. Is it a correct approach calculate the r^2 (r squared) for each pair of SNPs and then consider only the ones with a correlation smaller than 0.8 (I found this value as well used in literature)?

Answer 1

[Fabio Marroni]

I think it's the contrary. Usually, people reduce their dataset by REMOVING SNPs with r2>0.8. i.e. if you have two SNPs with r^2>0.8 you remove one of them.

[TitoPullo]

You're right, I made a mistake writing the post! Anyway is it a biologically correct approach?

[Fabio Marroni]

I never used this approach, but if you need to reduce the number of SNPs I think that this can be a reasonable approach, but this is just my personal opinion.

[TitoPullo]

Have you ever face with this problem (reduction of the number of SNPs)? If yes which approach did you used?

[Fabio Marroni]

No, I never had to reduce the number of SNPs. But I think that the most common approach is the one you described. So, my suggestion is to try that way.