starmie: Admixture models

In starmie: Population Structure Model Inference and Visualisation

knitr::opts_chunk$set(tidy = TRUE, 
                      fig.align = 'center', fig.width = 8, fig.height = 6)

In this vignette we use starmie to perform visualisation and inference on a sequence of ADMIXTURE models fit on the sample data supplied by the creators of ADMIXTURE.

Loading ADMIXTURE output and the 'admix' object

To begin we assume ADMIXTURE has been run on an example PED/BED file to analyse and there is a directory containing the resulting .P, .Q and logging files (if you want to perform inference).

To read a single ADMIXTURE run first supply the names of .P and .Q file and save it as an admix object using the loadAdmixture function.

library(starmie)
# get Q and P files for K = 3
p3 <- system.file("extdata/hapmap3_files", "hapmap3.3.P", package = "starmie")
q3 <- system.file("extdata/hapmap3_files", "hapmap3.3.Q", package = "starmie")

k3_admix <- loadAdmixture(q3, p3)
k3_admix

Optionally, if you would like to look at model fit statstics you can read in a log file as well. Now the admix object contains the estimated log-likelihood and cross-validation error.

k3_log <- system.file("extdata/hapmap3_files", "log3.out", package = "starmie")

k3_admix <- loadAdmixture(q3, p3, k3_log)
k3_admix

In general an admix object consists of the following elements about a single ADMIXTURE run:

list_names <- names(k3_admix)
list_description <- c("K parameter supplied to ADMIXTURE",
                      "Number of samples",
                      "Number of markers",
                      "Individual ancestral probability of membership to cluster",
                      "Estimated ancestral allele frequencies for each cluster",
                      "Model fit statistics")

knitr::kable(data.frame(attributes = list_names, 
                        description = list_description))

Constructing a barplot for a single admix object

The plotBar function works on admix objects and produces a facetted barplot in the same manner as for struct objects.

plotBar(k3_admix)

A regular barplot can be constructed by setting facet = FALSE.

plotBar(k3_admix, facet = FALSE)

Putting admix objects together the admixList object

If you have tried running ADMIXTURE with many different values of $K$ then you can use an admixList object to maninpulate them. To construct an admixList object use loadAdmixture for each pair of Q and P files and then pass the results to the admixList constructor function.

For example we have run ADMIXTURE on the sample data set with $K = 5$ using five-fold cross-validation.

admix_multi <- exampleAdmixture()
admix_multi

You can also plot multiple admix objects to see how cluster memberships change for different values of $K$.

plotMultiK(admix_multi)

Using ADMIXTURE output to perform inference.

The bestK method can be used to determine which value of K best explains the observed data using an elbow plot on the cross-validation error.

bestK(admix_multi)

References

Alexander, D. H., Novembre, J. & Lange, K. Fast model-based estimation of ancestry in unrelated individuals. Genome Res. 19, 1655–1664 (2009).

sessionInfo

sessionInfo()

starmie documentation built on May 1, 2019, 8:01 p.m.