Post date: Jun 15, 2017 8:1:9 PM
Compare CR between pheno and geno within a hypothesis.
pheno = geno + env noise. Pheno usually more correlated than geno values; env noise is correlated or natural selection.
can compare across groups within a hypothesis. More correlation within modules for many groups than others: genetic variants at all level different than subgroups or selection is favoring combos of modules within subgroups. Selection can be responsible for looser modularity in some groups (or different genetic architecture). We might be able to tease these part.
Teasing these apart: looked at individual snps so far…
are snps with high pips in 'all' high in subgroups. If so, if look different = selection
probably need to look at pips in sliding windows so that we are looking at snps associated with each other.
See position-modularity-results.xlsx for position results (pheno and geno).
15vi17.
position: I just read Klingenberg 2009 again. I can't just compare RV across hypotheses and use the one with the lowest RV coefficient for support. I have to also include the proportion of partitions for which the RV coefficient is less than or equal to the RV value for the partition of interest, which can be interpreted as the analog of such a p-value.
size: use resid-size-6vi17-subgroups.
can i just use a mantel test?
----
15vi17. So... we found that RV coefficients are subject to sample size and number of variables, so CR coefficient is the way to go. Paper: Adams, Evaluating modularity in morphometric data: challenges with the RV coefficient. R package: geomorph. function: modularity.test. Give it the matrix.
See geomoprh-modularity.txt
16vi17. Some notes after seeing the size pheno and geno results (size-modularity-results.xlsx):