I have association results for 15 SNPs, which I would like to replicate in a second study but only 3/15 SNPs are present in both studies. Although I could try imputing the missing SNPs, I want to see if there are any good proxy SNPs I could use as substitutes for the missing SNPs. My question is, what is an acceptable r2 threshold for a proxy SNP? Is r2>0.5 between a proxy SNP and a "lead" SNP acceptable? I guess the weaker the r2 the lower the power will be in the second study.
It is better to choose SNPs that are in absolute LD (r2 =1, D prime =1) for replication. If not, the general consensus among researchers is r2 should be at least 0.8 or more.
My advice would be that SNPs should be selected with a Minor Allele Frequency (MAF) above or, at least, equal to 5% for the study population (HapMap CEU). Tag SNPs should be selected with a correlation coefficient r2 = or ≥0.8.
Jose Rueff
As I understand from your question, you already genotyped the second set for the "new" SNPs. You basically haven't chosen your SNPs. I agree with Jimmy Liu that your effect size of your first associations will be very important: low effect sizes means that lower r2 are more problematic. Choosing SNPs that were in perfect LD would have been the ideal situation. r2
r2 is sensitive to allele frequency. So in case your SNPs are rare (MAF < 5% for example) then you may be better off using D'. In this scenario there may be tag-SNPs with an r2 < 0.8 consensus threshold; but they could be still valid tags if D'=1.
Actually, it depends on the effect size. If the effect size is large, than r2 can be taken smaller. You can chose several SNPs with different r2 and conduct sensitivity analysis. With r2>=0.8 you have quite small number of subjects carrying mismatched alleles from SNPs in LD (compound heterozygotes) so it is quite reasonable to expect that the effect is not entirely due to those heterozygotes.
- Badges
- Science topic
- Similar topics
- Biological Science
- Biochemistry
- Biomolecules
- Peptides