ancestry

this release adds an initial implementation of an ancestry sub-command that can use a set of labelled samples (with extracted somalier files) to train a small neural network which is then used to predict the ancestry of incoming samples.
the implementation is incomplete, but works for well-behaved data. Here is an example:
http://home.chpc.utah.edu/~u6000771/somalier/somalier-ancestry.n.html

This is possible thanks to a very fast randmized PCA implementation (along with a neural network framework) from @mratsim in Arraymancer.

There are also improvements for huge cohorts. See below for full change-set.

Installation

grab the static binary below, or use docker via brentp/somalier:v0.2.7

v0.2.7

• new subcommand ancestry to predict ancestry using a simple neural network on the somalier
  sketches. creates an interactive html output and a text file
• fix for "Argument list too long" on huge cohorts (#37)
• sub-sample .pairs.tsv output for huge cohorts -- only for unrelated samples.
• better sub-sampling of html output

sites files (unchanged from previous releases)

These sites files are build-specific, but as of this release, once the sites are extracted, the resulting somalier files can be used to compare samples even across genome builds.

sites.hg19.vcf.gz
sites.hg38.nochr.vcf.gz
sites.GRCh37.vcf.gz
sites.hg38.vcf.gz