torchtree-physher

About torchtree-physher

torchtree-physher is a python package providing fast gradient calculation implemented in physher for torchtree.

Getting Started

A C++ compiler such as g++ or clang++ is required. On Debian-based systems, this can be installed via apt:

sudo apt install g++

On MacOS, it is recommended to use the latest version of clang++:

brew install llvm

The pybind11 library is also used for binding the C++ functionality to Python.

Dependencies

• physher
• pybind11
• PyTorch
• torchtree

physher is a phylogenetic program written in C that provides C++ wrappers to compute the tree and coalescent likelihoods and their gradients under different models.

To build physher from source you can run

git clone https://github.com/4ment/physher
cmake -S physher/ -B physher/build -DBUILD_CPP_WRAPPER=on -DBUILD_TESTING=on
cmake --build physher/build/ --target install

Check it works (optional)

ctest --test-dir physher/build/

Installation

To build torchtree-physher from source you can run

git clone https://github.com/4ment/torchtree-physher
pip install torchtree-physher/

Check install

If the installation was successful, this command should print the version of the torchtree_physher library

python -c "import torchtree_physher;print(torchtree_physher.__version__)"

Command line arguments

The torchtree-physher plugin adds these arguments to the torchtree CLI:

torchtree-cli advi --help
  ...
  --physher             use physher
  --physher_include_jacobian
                        include Jacobian of the node height transform in the node height gradient
  --physher_disable_sse
                        disable SSE in physher
  --physher_disable_coalescent
                        disable coalescent calculation by physher
  --physher_site {weibull,gamma}
                        distribution for rate heterogeneity across sites

Features

Tree likelihood

Some types in the JSON configuration file have to be replaced in order to use the tree likelihood implementation of physher. You simply need to add torchtree_physher. before a model type. Here is a list of models implemented in this plugin:

• TreeLikelihoodModel
• Substitution models:
  • JC69
  • HKY
  • GTR
  • GeneralNonSymmetricSubstitutionModel
• Tree models:
  • UnRootedTreeModel
  • ReparameterizedTimeTreeModel
• Clock models (optional):
  • StrictClockModel
  • SimpleClockModel
• Site models:
  • ConstantSiteModel
  • GammaSiteModel
  • InvariantSiteModel
  • WeibullSiteModel

Note that the type of every sub-model of the tree likelihood object (e.g. site, tree models...) has to be replaced.

For example if we want to use ADVI with an unrooted tree and a Weibull site model:

torchtree-cli advi -i data.fa -t data.tree -C 4 > data.json
sed -i -E 's/TreeLikelihoodModel/torchtree_physher.TreeLikelihoodModel/; s/UnRootedTreeModel/torchtree_physher.UnRootedTreeModel/; s/WeibullSiteModel/torchtree_physher.WeibullSiteModel/' data.json
torchtree data.json

The JSON file can be created directly using the --physher option:

torchtree-cli advi -i data.fa -t data.tree -C 4 --physher > data.json

Coalescent models

Here is a list of coalescent models implemented in this plugin:

• ConstantCoalescentModel
• PiecewiseConstantCoalescentGridModel (aka skygrid)
• PiecewiseConstantCoalescentModel (aka skyride)
• PiecewiseLinearCoalescentGridModel

License

Distributed under the GPLv3 License. See LICENSE for more information.

Acknowledgements

torchtree-physher makes use of the following libraries and tools, which are under their own respective licenses:

• physher
• pybind11
• PyTorch
• torchtree