Visualizing geographic structure and demographic history: (very) short course

This is the material for a short (3 hour) course I taught at the UCLA/La Kretz Workshop in Conservation Genomics, 22-27 March, 2015.

Goals/skills:

1. Describe different causes and patterns of population structure.
2. Use principal components analysis (PCA) to visualize population structure.
3. Use BEDASSLE to fit models of isolation by distance and/or environment.

Incidental skills:

1. Interpreting the output of a Markov chain Monte Carlo (MCMC) algorithm.

Prerequisites:

1. Install R

2. and the R package BEDASSLE:

   install.packages("BEDASSLE")

To get the data used in the examples, download and uncompress this tarball to the folder you've put this git repository in.

In this repository

1. The presentation.
2. Example of running PCA on some data.
3. Example of running BEDASSLE on some simulated data.
4. Example of running BEDASSLE on two real datasets: teosinte and the HGDP.

Each of the above html files are made from the corresponding Rmd files of the same name; you can read the source code to see what happens behind the scenes when the R code isn't in the presentation.

Outline

1. PCA and visualization (45min)
   • Goal: low-dimensional summary.
   • In practice: examples from the literature
   • Properties of PCA
     • direction of maximum variation: weights on alleles
     • decomposition of variance
     • matrix decompositions
     • spatial structure and Fourier modes
   • Technical issues
     • normalization
     • weighting and sample sizes
     • correlated markers
2. Doing PCA: hands-on (45min)
   • Three populations, simulated
     • computing the covariance matrix
     • PCA
   • POPRES (as in http://www.ncbi.nlm.nih.gov/pubmed/18758442)
     • covariance matrix provided
     • downweighting populations
     • subsets of the genome
3. Continuous geography (45min)
   • Nearby things are more similar than distant ones
     • ... because they are more closely related.
     • Migration, coalescence, covariance, and genetic distance.
   • Isolation by distance/environment/ecology/etcetera
     • resistance distance
   • Phenomenological model: correlated allele frequencies
4. Using BEDASSLE (45min)
   • BEDASSLE's parameterization
   • Lightning introduction to MCMC
     • acceptance rates
     • likelihood profile
     • mixing and stationarity
   • Simulated data

License and reuse

This work, including the code and presentation, is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.