knitr::opts_chunk$set(fig.width = 5, fig.height = 5)


This vignette shows how to use the R package disclapmix that implements the method described in [@AndersenDisclap2013]. For a more gentle introduction to the method, refer to [@AndersenDisclapIntroduction2013].


A Danish reference database [@Hallenberg2005YchromosomeSH] with $n = 185$ observations (male Y-STR haplotypes) at $r=10$ loci is available in the danes dataset. Let us load the package as well as the data:



The database is in compact format, i.e. one unique haplotype per row. To fit the model, we need one observation per row. This is done for example like this:

db <- as.matrix(danes[rep(seq_len(nrow(danes)), danes$n), seq_len(ncol(danes)-1)])

Also, note that the database is now an integer matrix.

To fit a model using 2 clusters, the following command can be used (note the L postfix to emphasize that the number is an integer):

fit <- disclapmix(x = db, clusters = 2L)

The number of clusters is not known beforehand. Here, the numbers 1 through 5 are tried and the best one according to the BIC criteria [@BIC] is taken:

clusters <- 1L:5L
fits <- lapply(clusters, function(clusters) {
  fit <- disclapmix(x = db, clusters = clusters)

marginalBICs <- sapply(fits, function(fit) {

bestfit <- fits[[which.min(marginalBICs)]]

The best fit is now in the bestfit that can be inspected by print (default method called when the variable is just written) or summary which (currently) give the same output:


We can also plot the fitted model:


There are important information returned by disclapmix, e.g. the central haplotypes and the dispersion parameters for the discrete Laplace distributions:


The returned object is described in ?disclapmix and objects can be inspected using e.g. str().

Haplotype frequencies can be obtained using the predict function. Note, that this is done per haplotype (danes) and not per observation (db):

disclap_estimates <- predict(bestfit, 
                             newdata = as.matrix(danes[, 1:(ncol(danes) - 1)]))

These can be compared to the database frequencies:

ggplot() +
  geom_abline(intercept = 0, slope = 1) +
  geom_point(aes(x = danes$n/sum(danes$n), y = disclap_estimates)) +
  labs(x = "Observed frequency",
       y = "Predicted frequency (discrete Laplace)") +
  theme_bw() +
  scale_x_log10() +


Try the disclapmix package in your browser

Any scripts or data that you put into this service are public.

disclapmix documentation built on May 1, 2019, 8:49 p.m.