Structure
- continued from yesterday ...
Things I learned from chatting with Rulon
- standard is to run a few iterations of each value of K
- do NOT unclick "Compute probability of data (for estimating K)" ... this generates the Ln P(D) statistic that is reported in publications
- can average the Ln P(D) values for each value of K ... the most positive value of Ln P(D) is the K that Structure likes best
- averaging the 3 iterations of each value of K ... K=1 gives most positive Ln P(D) ... :(
GENEPOP
- Option 1 - Hardy-Weinberg exact test
- sub-option 3 - exact tests
- Option 2 - Linkage disequilibrium
- sub-option 1 - each pair of loci, log likelihood ratio
- Option 3 - Population Differentiation
- sub-option 3 - genotypic differentiation for all populations
- Option 5 - Basic Information
- sub-option 1 - basic info (genotypic matrices, number observed/expected homozygoes/heterozygotes, allele frequencies)
- saved all output to GenePop folder
- using Riley et al 2006 as framework for which analyses to run
- they report (in supplement, Table S2) observed and expected heterozygosities for each locus as well as overall
- where do they get that from GenePop?
- looks like they gotta calculate it themselves:
- Ho = (number observed heterozygotes) / (number observed heterozygotes + number observed homozygotes)
- Ho overall = average of each locus Ho
- but don't know how they get an 'overall' for each locus - it's not the average
Things I learned from chatting with Rulon
- try FSTAT ... see if that'll give you Ho straight up
|