Partis can be installed either with Docker or from scratch.
Using Docker makes the installation process much easier, because it installs specific versions of each dependency in a controlled environment. The process of running, however, is then less convenient since you're inside Docker (the use of batch systems, for instance, will be difficult).
Installing without Docker, on the other hand, will require more effort to install, but once installed it'll be easier to use. The closer your system is to the latest Debian/Ubuntu, the easier this process will be (RHEL variants and macOS typically work without too much trouble).
If you want to make simulated samples, you'll also need to install some R packages.
Docker can be installed using the instructions here. If you're new to Docker, start with the official quick start guide. Then,
sudo docker pull quay.io/matsengrp/partis
sudo docker run -it --name container-1 -v ~:/host/home quay.io/matsengrp/partis /bin/bash
The sudo may not be necessary for some systems. With docker run, we create a new container from (i.e. instance of) the partis image. The -v mounts your home directory on the host machine to the path /host/home inside the container, so we can easily extract results.
Docker's default escape key is ctrl-p, which unfortunately conflicts with common command line shortcuts. To change this put{"detachKeys": "ctrl-q,q"} into the file ~/.docker/config.json which will swap it to ctrl-q.
When finished, you don't want to just exit and docker run again -- this will create a new container.
One solution is to run Docker inside tmux in which case you can just leave the container open.
Alternatively, you can detach from the container with ctrl-q q (or whatever you've remapped ctrl-p q to).
To then reattach to this running container, run docker attach container-1.
If you don't plan on reattaching to a container, you should run with --rm so it is removed when you exit; otherwise you'll use up a lot of disk space with accumulated old containers (view with sudo docker ps -a).
To install without Docker, you basically need to run the commands in the Dockerfile, starting by installing micromamba (essentially an alternate flavor of conda).
You can probably ignore the $MAMBA_USER stuff.
Because mamba/conda and pip do not play well with each other, if you've used pip in the past but are going to now use conda, and you won't need pip in the future, you should completely remove ~/.local.
If you might need pip in the future, you should expect some difficulty with having both conda and pip on the same system.
In most cases you can prevent conda from finding the packages in ~/.local (and consequently breaking) by setting export PYTHONNOUSERSITE=True (see this issue for some context).
You may also need to unset LD_LIBRARY_PATH.
If you're on a debian variant, run the apt-get install as written.
On other distros, or on macOS, you'll have to figure out the equivalent package names, but after that just swap yum/brew for apt-get.
In place of COPY, you should clone the partis repo and init/update its submodules:
git clone git@github.com:psathyrella/partis
git submodule update --init --recursive
ARG lines should be export commands, and substitute the line WORKDIR /partis with cd partis/.
In order to avoid polluting your environment, we do not automatically add partis to your path. Several methods of accomplishing this are described here.
By default, partis simulation generates trees using the R packages TreeSim and TreeSimGM, so if not already present they'll need to be installed (along with R).
Alternatively, if you have a list of your own newick trees in a file you can pass this in with --input-simulation-treefname, in which case you can avoid R installation entirely.
One way to install R on debian/ubuntu (and thus also within a partis container) is:
apt-get install -y dirmngr apt-transport-https ca-certificates software-properties-common gnupg2
# UPDATE maybe should be keyserver.ubuntu.com?
apt-key adv --keyserver keys.gnupg.net --recv-key 'E19F5F87128899B192B1A2C2AD5F960A256A04AF' # this R installation stuff was in the Dockerfile for a while, but this line was crashing on dockerhub, and it appears that keyservers are just too flakey to use in docker files
add-apt-repository 'deb https://cloud.r-project.org/bin/linux/debian stretch-cran35/' # this will have to change when anaconda changes debian releases, but the tree sim packages require a more recent version of r than is in the debian repos
apt-get update && apt-get install -y r-base
# UPDATE maybe should have this here somewhere https://mirror.las.iastate.edu/CRAN/
Note that TreeSim/TreeSimGM may require a more recent R version than is in the default repos (check with apt get show r-base-core, then compare to requirements in TreeSim/TreeSimGM cran pages).
Note also that if you instead install R with conda, it replaces your entire compiler toolchain (e.g. screws up where your system looks for gcc) so you probably won't be able to compile anything afterwards.
You can then install the required R packages with the interactive R command:
> install.packages(c("TreeSim", "TreeSimGM", "geiger", "MASS"), repos="http://cran.rstudio.com/", dependencies=TRUE)
If necessary, you will probably want to follow the prompts for installing to a personal (rather than system-wide) library location. If you instead install from the command line with:
R --vanilla --slave -e 'install.packages(c("TreeSim", "TreeSimGM", "geiger", "MASS"), repos="http://cran.rstudio.com/", dependencies=TRUE)'
You may need to run with sudo -i in front, and the resulting library permissions may cause problems.
The default mutation model also requires compilation of an updated (development) version of bpp that's in packages/bpp-newlik/:
./bin/build.sh with-simulation # this takes a while, maybe 20-60 minutes
However, if you don't much care about mutation model accuracy (in particular, if you don't care about modeling different rates between different bases e.g. A->C vs A->T), you can set --no-per-base-mutation and it'll fall back to an older, already-compiled bpp version in packages/bpp.
In order to make the MDS plots, you need R and the bios2mds package installed:
apt-get install xorg libx11-dev libglu1-mesa-dev r-cran-rgl # may not be necessary, depending on existing R install, but is at least necessary in docker
conda install -y -cr r-rgl r-essentials
unset R_LIBS_SITE
R --vanilla --slave -e 'install.packages(c("class", "bios2mds"), dependencies=T, repos="http://cran.rstudio.com/")' # this may be all the deps: c("picante", "pspline", "deSolve", "igraph", "TESS", "fpc", "pvclust", "corpcor", "phytools", "mvMORPH", "geiger", "mvtnorm", "glassoFast", "Rmpfr")