Phylogeny inference using deep neural network

Setting up experiment environment

Here are the recommended steps to set up the experiment environment:

1. Install Anaconda or Miniconda.
2. Under the root folder of this project, create a conda environment by running conda env create -p env/ -f environment.yml.
3. Execute conda activate env/ to activate the conda environment. Do not omit the trailing slash.
4. Run ./setup.py develop to install evosimz for tree simulation.
5. Create folders data/trees and data/models under the root folder of the project.

Now you can run files under the "bin" folder for experiments.

To quit the environment, run conda deactivate.

To reactivate the environment in the future, run conda activate env/. There is no need to run ./setup.py develop unless files in the evosimz module are changed.

Simulating trees

To simulate quartets (phylogenies with four taxa):

Run evosimz quartet <simulator> <dataset name> <sample size> <job count>.

simulator can be either the path to a pickled simulator or a simulator variable defined in any of the modules in evosimz/simulators/.

dataset name will be the folder name. The generated sample folder will be placed under data/trees.

Please refer to data/scripts/bash_scripts/sim_train.sh for example simulation commands.

Training and prediction

Run scripts under bin/ to train and predict. Take bin/q10 as an example:

bin/q10 train lists all arguments for training.

bin/q10 predict lists all arguments for prediction.

Models will be saved under data/models/<script name>.

Please refer to data/scripts/bash_scripts/sim_train.sh for example training commands, and data/scripts/bash_scripts/prediction.sh for example prediction commands.