SINGER

Warning

SINGER support is still experimental, and our integration is not as well tested or supported as tsinfer or Relate. On the models we have tested, using mrpast with SINGER was not as accurate as tsinfer+tsdate. It may be worth comparing multiple ARG inference methods on your model(s).

This walkthrough uses this fork of SINGER that has a few minor changes from the main fork.

Installation:

git clone https://github.com/dcdehaas/SINGER.git
cd SINGER
mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make -j 4
cd ..
./make_release.sh
export PATH=$PATH:$PWD/releases/singer-0.1.8-beta-linux-x86_64/

After running the above build and setting $PATH to include the SINGER executables you should have everything needed to run SINGER. You only need to do the build once, but you’ll need to set $PATH everytime you want to use SINGER (or add it to your login script).

Steps 1-2: Same as Relate

See the Relate walkthrough and do steps 1 and 2, which should result in some VCF files and population map JSON files.

Step 3: Infer ARGs with SINGER

Now we can run SINGER on our data.

mkdir -p 5d1e.singer/
mrpast arginfer -j 2 --mut-rate 1.2e-8 --recomb-rate ratemap_Hg38_chr1.txt --tool singer 5d1e.simdata/5d1e_msprime_ 5d1e.relate/5d1e_rel_ 5d1e.simdata/5d1e_msprime__0-0.trees.popmap.json

Note

If you get error RuntimeError: Could not find executable parallel_singer then you don’t have SINGER setup properly on your $PATH. See installation instructions above.

Step 4-5: Same as Relate

See the Relate walkthrough and do steps 4 and 5, except instead of 5d1e.relate use 5d1e.singer.

Note

SINGER performs MC/MC on both the topology of the ARGs and the branch lengths. You may need to increase the number of samples for mrpast arginfer (see --samples) to get an adequate exploration of the ARG space.