I'm working on a fish species with populations distributed across a river network, using microsatellite markers to investigate population genetics and landscape genetics in a conservation context. I would like to:
1) Tease apart which landscape-based distance variables (distance between, elevation change, count of anthropogenic structures, etc.) are contributing most to fragmentation of identified populations; and
2) If the top contributing types of features are anthropogenic (e.g., dams or road crossings), I would like to identify the individual features within that type that are functioning as the most restrictive barriers.
Any ideas of potentially useful programs, analyses, or references would be appreciated. Thank you.
I would suggest looking at the new Bedassle package in R, from Gideon Bradburg, Peter Ralph, and Graham Coop. It does many of the things you are looking to. It uses a fairly complex Bayesian framework, and getting it running in R is not trivial. But it might outperforming other platforms like genalex, spagedi, arlequin, or genodive, that will also do some of the analyses you want.
Great question, Andrew! I'll be interested in seeing what you use.
It really depends what your data looks like, and what statistics you are looking to use for comparing landscape variables with genetic and geographic distance. I just included a short list here based on the most common methods I have seen, but there are lots more.
If you are looking to partial geographic (or riverine) distance out, the most common statistics are Mantel tests (typically performed in the R packages ecodist or vegan) and multivariate stats such as redundancy analysis or distance-based redundancy analysis (usually performed in vegan again). Partial mantel tests are subject of debate over their precision and accuracy, so be sure to read up on that debate if you use them.
Riverine systems also lend themselves well to graph theoretic techniques (see Garroway et al. 2008) and network analyses. These types of analyses are often performed in the R package igraph or the specialized genetics programs mentioned in Garroway. I typically use igraph though, does the same thing.
I also like to use spatial regressions and mixed models, but these techniques depend on your genetic variables being independent or including an effect for repeated sampling. Spatial regressions can be found in spdep (R package) and mixed models can be performed in many R packages, I prefer lme4.
There are many other techniques such as random forests and Bedassle, but again, which technique you use depends on your data, and what variables you ultimately derive (i.e., how do you represent landscape variables and choice of genetic variable). Hope this helps.
Liz
The effects of fragmentation on the genetic diversity vary among life forms. In plants it is estimated that up to 100 years after detected (Molecular Ecology, 17, 5177-5188). Fish depend on the generation time.
Thank you all for the suggestions thus far - they've been very helpful. Bedassle certainly appears promising! Geneland also appears to address some of the questions I had in mind, although I'm not entirely sure how to represent my environmental data within the program upon first read through the user guide. I also recently came across BIMr by Faubet (2007). Reading through this user guide, it looks like BIMr may also be a relatively simple way to address some of my questions. Judging by citations, it doesn't appear as though BIMr has been widely applied...does anyone have any experience with this program, in particular?
You can use the R package riverdist to calculate pairwise water-way distance between sample sites for IBD. The adespatial package is great, and you might want to consider Bayesass to estimate migration between populations.
Thanks Spencer Bruce. I'll have to look into riverdist - that certainly sounds useful for my applications in stream systems.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology