来自researchgate

Jonathan Marescaux：

Why do Fst values from Arlequin differ from other software?

I was looking for a software which gives the Fst p-value and when I analyzed the results obtained with Arlequin, I realized that the Fst values were different than those from Spagedi or Genepop ...

Wael Yakti：

Fst values could be estimated using different methods ( different mathematical equations) . The first one was proposed by Wright (1951), and later came Cockerham (1973) and Weir & Cockerham (1984) ... etc,

sometimes different software use different methods, and thats what you can find out in their manuals ... nevertheless, they all serve the same purpose and can be trusted for your comparisons.

Someone：

Hi Jonathan,

Wael is correct that there are multiple ways to calculate Fst. However, your choice of which version of Fst is actually very important, since each method has different behaviour under different conditions (e.g. sample size, missing data, locus polymorphism etc.).

Genepop has its own way of calculating Fst across loci. This methods differs slightly from that proposed by Weir & Cockerham, (1984). In my R package, diveRsity, I calculate Fst exactly as specified in Weir & Cockerham, (1984) and when I compare the result to those returned by genepop, they are typically only different at the 3rd-4th decimal place. Whereas locus Fst's are identical. This difference in overall Fst is as a result of the way that diveRsity and genepop weight locus estimates, but it is never likely to effect experiments conclusions, unless the variance in missing data across loci is very large, perhaps.

If the differences between Arlequin and the other software are larger than 3-4 decimal places, then there may be a more meaningful difference in the way the softwares are calculating the statistic. You should check the arlequin manual to see which Fst version it is calculating for you (Maybe Slatkin 1995 or Phi_st etc.) The pdf at the link below provides some information about different ways to calculate F-statistics (http://www.library.auckland.ac.nz/subject-guides/bio/pdfs/733Pop-g-stats2.pdf).

Which marker type are you using? If it is highly polymorphic microsatellites, then perhaps you should consider Jost's D, especially if you are only interested in genetic differentiation. Also you should consider using 95% confidence intervals for hypothesis tesing. P-values in this context have a very high type I error rate.