Hi there.
I am currently using microsatellites to study the genetic structure of a haploid plant pathogen. i have measured Fst between populations in a pairwise manner using arlequin (with 10000 permutations) in two ways. firstly using my SSRs in the traditional sense, using all of their alleles (recoded as allele size after binning in genemapper), and secondly by recoding the data as binary, using the SSRs simply as presence absence markers (absence being a failure to amplify or null allele), given the chromosomes of my pathogen are highly unstable and failure to amplify is frequent.
In my pairwise analysis i find that significant levels of differentiation (Fst) are driven up by specific SSR loci, and differentiation is especially high when using the SSRs as presence absence markers, with the same loci driving this result. to look more in detail at this i decided to look at the overall Fst measures for each SSR locus using the hierfstat R package. As expected the loci that appear to drive the pairwise Fst measures up in arlequin have the highest overall Fst value, calculated by the varcomp function (as a presence absence marker with two alleles), however when i test the significance of this using the test.between function the result is non-significant (p = 0.112). Also using the same test on a marker with a lower Fst value (coded as normal with multiple alleles) yields a significant result (p = 0.001). what could be causing this? is the permutation test for significance in this instance unsuitable for a marker with only two alleles? and if so is there some kind of adjustment i can make to properly test the significance of Fst for these markers when they are coded as such?
I am not familiar with the latter packages, but I would suggest that you try to understand how the test of significance is conducted. Higher FST does not necessarily mean more significant in terms of P-value. Usually, the significance of Fst is determined by some sort of resampling procedure (such as bootstrapping or jackknifing) that allows you to put a confidence interval around the point estimate...or to test if your estimate is greater than (or lower than) would be expected if the data were randomized over individuals in different populations.
Using some sort of weighting scheme based on your level of polymorphism might be useful in terms of improving the accuracy of your estimate, but the test of significance itself is has nothing to do with polymorphism and is more affected by the number of loci you are using or the number of populations that you are examining (depending on how you resample).
- Badges
- Science topic
- Similar topics
- Biological Science
- Genetics
- Genetic Analysis