Peter Pfaffelhuber, Franziska Grundner-Culemann, Veronika Lipphardt, Franz Baumdicker

Overview:

AIMsetfinder is a collection of Rscripts to identify sets of Ancestry Informative Markers (AIMs), that minimize the logloss error of a naive Bayes classifier.

It takes as input:

• SNP data (e.g. 1000 Genomes SNP data or user's own data in vcf.gz format)
• biogeographic information (or alternatively any discrete phenotype)

to select a set of specified size of optimal AIMs to classify the samples.

The output is:

• a vcf.gz file with the selected AIMs
• a list of SNP identifiers

which can be used in ancestry inferrence methods.

• Furthermore the posterior probabilites of a naive classifier based on these AIMs are given for the input data.

• Quick start
• Dependencies
• How to run
• Directory structure and analysis output
• Command line arguments

git clone https://github.com/fbaumdicker/AIMsetfinder.git
cd AIMsetfinder

Install dependencies and then run the test: Rscript pipeline_example.r

For data analysis as well as for our simulation studies, we rely on R scripts.

• For multicore-computing, we require the R-package parallel.
• Since, both data from the 1000 genomes project, which is analysed here, and the coalescent simulations, come in vcf-format, we require the R-package vcfR.
• For some steps in the analysis, we use vcftools \cite{Danecek2011} and bcftools \cite{Li2011}, both can be installed using

sudo apt-get install vcftools bcftools

In addition, data from the 1000 genomes project (phase 3) was downloaded, as well as information on the sampling locations. For this, we used

wget ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/release/20130502/ALL.*
wget ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/release/20130502/integrated_call_samples_v3.20130502.ALL.panel

The latter file was renamed {\tt 1000G_SampleListWithLocations.txt}, and the first row (the header) was removed.

The simulation studies are performed using msprime. Most easily installed via pip3 install msprime. Msprime is a fast coalescent simulator. In particular, structured populations (with varying population sizes etc) can be simulated. We are using the python-interface of msprime. See msprime documentation for more information.

• msprime
• vcftools
• bcftools
• nnet
• vcfR
• parallel R package (part of base R)

To run the test set: Rscript example_pipeline.r

In data/sim/ooa/, you will find the file ooa_chromosome_1_example.vcf.gz, a small set of 240 simulated individuals that is used in this tutorial.

The analysis generates the following files:

./AIMs                list of identifiers of the chosen AIMs
./AIMs.vcf.gz         corresponding states for all individuals
./predictions.csv     table of posterior probabilites for all classes/BGAs as predicted by naive Bayes using AIMs
./classifications.tab classification into the class with the largest probability as in predictions.csv 

In which step different files are produced is described in more details in readme.pdf.