I created an algorithm to generate SNPs for random people of different descents - based on HapMap data. While this works good, there is something else I want to take into consideration. So if a SNP has a MAF of 0.01, with an LD of 1, the generated results will, in most cases be sufficient. This is not the case for a MAF of 0.4 and LD of 0.8.

The HapMap LD files have, besides the LD data, also LOD and D-prime. The question is, is there a set of parameters so that as much of the relevant LDs are filtered out of the total set?

I want to filter this out, since, for example, there are LDs described with a value of 0, or an LOD < 1, and problematic D-primes.

Trying to rephrase the question:

Given a huge set of LD information (250GB), which are not filtered on statistical power, I want to make a subselection of LD information that are 'relatively' descriptive of the link with respect to another SNP. So I need to have a set of parameters to filter the LD information from HapMap. Of which possible values are LD, LOD, and D-prime.

Eyeballing the LD dataset, it doesn't gives a good instinctive boundary. A problem with using the data as is, is that both the D' as LOD can be both high and low. I know the parameters have an interdependence, but not a one-to-one mapping.

Does anybody have experience in filtering non-descriptive LD values from (for instance) the HapMap LD dataset?

  • $\begingroup$ I don't get your question. LD, LOD and D' are not independent parameters but scores describing the heritability linkage between SNPs. For GWAS SNPs are typically pruned at an LD linear $r^2$>0.8 to ensure independence but I don't think this is really relevant for the simulation you are doing. Also I don't see why a LD value of 0 is problematic, this means that two SNPs are completely independent. $\endgroup$ – cagliari2005 Jul 2 '15 at 19:27
  • $\begingroup$ @cagliari2005 I know what LD means, but working with lower LDs make generating alleles for SNPs that were generated before harder, since the MAFs differ. Wrt D-prime, values of -1 and 1 seem good, but only if either the MAFs are similar, or the population size is big enough. LOD of 3 seem to make sense, but for certain purposes possibly to restrictive $\endgroup$ – Danielson Jul 3 '15 at 6:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.