Probabilistic graph models for landscape genetics
- Published
- Accepted
- Subject Areas
- Computational Biology, Scientific Computing and Simulation, Spatial and Geographic Information Systems
- Keywords
- Landscape genetics, Graph models, Bayesian inference, Open source software
- Copyright
- © 2016 Milligan
- Licence
- This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, reproduction and adaptation in any medium and for any purpose provided that it is properly attributed. For attribution, the original author(s), title, publication source (PeerJ Preprints) and either DOI or URL of the article must be cited.
- Cite this article
- 2016. Probabilistic graph models for landscape genetics. PeerJ Preprints 4:e2225v2 https://doi.org/10.7287/peerj.preprints.2225v2
Abstract
Progress in landscape genetics depends on a strong conceptual foundation and the means of identifying mechanistic connections between environmental factors, landscape features, and genetic or genomic variation. Many existing approaches and much of the software commonly in use was developed for population genetics or statistics and is not entirely appropriate for landscape genetics. Probabilistic graph models provide a statistically rigorous and flexible means of constructing models directly applicable to landscape genetics. Sophisticated software also exists for the analysis of graph models. However, much of that software does not handle the types of data used for landscape genetics, model structures involving autoregressive spatial interaction between variables, or the scale of landscape genetics problems. Thus, an important priority for the field is to develop suitably flexible software tools for graph models that overcome these problems and allow landscape geneticists to explore meaningfully mechanistic and flexible models. We are developing such a library and applying it to examples in landscape genetics.
Author Comment
The reviewer's comments focused on four topics. First, additional references were desired concerning the importance of mechanistic models. These have been added. Second, greater detail regarding analysis using graph models is desired, perhaps contrasting that with ``traditional'' approaches. I fully agree, but this might distract from the point. For one thing, there is no such thing as a ``standard'' analysis, so what would the comparison be? More generally, this request is the foundation of an entire paper. I am open to suggestions, but I'm not yet convinced that it would be a welcome addition to this particular paper. Nevertheless, I will add to the paper concrete examples of the graph models used by existing landscape genetics software. Additionally, I will point out potential modifications of those graphs to give a flavor of the power of manipulating graphs rather than being limited by specific models encoded in software. This is all well beyond the scope of this summary for the proceedings, but will be included within the resulting paper itself. Third, better understanding of the software under development is desired, especially at Perugia. I will include as much as possible in the talk; however, it is probably premature for the paper. I feel that the concrete examples of graphs mentioned previously will greatly improve the understanding of this point as well. Finally, the reviewer focuses on the difference between population- and individual-based analyses. I will add a section to the paper that clarifies how this difference is manifest in graph models and how that difference is not nearly as fundamental as it might appear from traditional statistical analyses. Beyond these points, I have reduced the length of this summary to fit within the guidelines of the proceedings. The full paper will use additional space to reintroduce the table of software and introduce a series of figures as described previously.