Post date: Nov 19, 2013 8:48:59 PM
We asked whether parallel divergence change SNPs (the top 99.5th quantile) from the experiment were in parallel high genetic differentiation regions (that is high genetic differentiation regions identified by the HMM in two or more populations) more often then expected by chance (i.e., under the null hypothesis of no association between these metrics). We addressed this question with a permutation test (1000 permutations). The results are below.
99.5th quantile parallel divergent change and parallel high genetic differentiation
Thus, we have evidence that the top (99.5th quantile) parallel divergent change SNPs in the experiment occur significantly more often in parallel (2 or more pairs) high genetic differentiation regions than expected by chance, and also evidence that these SNPs occur in high genetic differentiation regions between R12A and R12C significantly more often then expected by chance.
The new R code for this analysis is below, and an R workspace with these specific permutation results is in projects/timema_wgwild/experiment/experq99.5.rwsp. Finally, whereas there is a similar trend with the 99.9th and 99.0th quantile these are not significant. Thus, the trend, but not nominal statistical significance, is robust to different quantile cutoffs.
## combined evidence across all four population pairs
cnt<-apply(hmmStates[X,]==1,1,sum,na.rm=TRUE)
obsCnts<-numeric(5)
for(j in 1:5){
obsCnts[j]<-sum(cnt==(j-1))
}
nsam<-1000
Nx<-length(X)
nullCnts<-matrix(NA,nrow=nsam,ncol=5)
for(i in 1:nsam){
Y<-sample(1:nl,Nx,replace=FALSE)
cnt<-apply(hmmStates[Y,] == 1,1,sum,na.rm=TRUE)
for(j in 1:5){
nullCnts[i,j]<-sum(cnt==(j-1))
}
}
obsGr1<-sum(obsCnts[3:5])
nullGr1<-apply(nullCnts[,3:5],1,sum)
mean(nullGr1 >= obsGr1)