phission

phission: phasing via nuclear "fission"

This package implements haplotype phasing by formulating it as a low-rank matrix completion problem. phission uses minimum nuclear norm as a convex relaxation of low-rank.

Installation

Clone this repo

git clone https://github.com/captaincapsaicin/phission.git

From the project root, create a virtual environment using virtualenv

cd phission
virtualenv venv

Activate the virtual environment

source venv/bin/activate

Install numpy before msprime (msprime installation depends on it)

pip install numpy --no-cache-dir
pip install msprime --no-cache-dir

Then install the rest of the dependencies (this may take some time)

pip install -r requirements.txt

Running the example

python run_phission.py --num-haps 100 --num-snps 60 --seed 1

For help with command line options, run:

python run_phission.py -h

Running beagle for comparison

Make sure you have the beagle jar in your working directory

wget https://faculty.washington.edu/browning/beagle/beagle.27Jan18.7e1.jar

Then run the script

python run_beagle.py --num-haps 100 --num-snps 60 --seed 1

Running a suite of experiments

To run a series of experiments (to collect some statistics on performance), run

python run_experiments.py --num-experiments 8 --recombination-rate 0.0 --haps-snps 10,10 20,20

Which writes the statistics to a pickle file, which you can then examine with:

import pickle

import numpy as np
from tabulate import tabulate


def get_stats_back(num_haps_snps_list):
    with open('beagle_stats.pkl', 'rb') as f:
        beagle_stats = pickle.load(f)
    with open('phission_stats.pkl', 'rb') as f:
        phission_stats = pickle.load(f)

    for haps_snps in num_haps_snps_list:
        print()
        headers = [str(haps_snps), 'rank reduction', 'percent switch error']
        data = [['phission', np.mean(phission_stats[haps_snps]['rank_true'] - phission_stats[haps_snps]['rank_phased']), np.mean(phission_stats[haps_snps]['switch_error'] / phission_stats[haps_snps]['positions_phased'])],
                ['beagle', np.mean(beagle_stats[haps_snps]['rank_true'] - beagle_stats[haps_snps]['rank_phased']), np.mean(beagle_stats[haps_snps]['switch_error'] / beagle_stats[haps_snps]['positions_phased'])]]
        print(tabulate(data, headers=headers))

Followed by

In [7]: num_haps_snps_list = [(10, 10), (20, 20)]

In [8]: get_stats_back(num_haps_snps_list)
(10, 10)      rank reduction    percent switch error
----------  ----------------  ----------------------
phission               0.625                0.34059
beagle                -0.375                0.165035

(20, 20)      rank reduction    percent switch error
----------  ----------------  ----------------------
phission               0.125                0.306674
beagle                -0.625                0.101885

Running tests

py.test

Future work

Parallelize optimization using ADMM, performing phasing over overlapping windows, and then sensibly combining their results.

Advanced Notes

phission is compatible with Python 2 and 3.

If you want to put ipdb hooks in the code, remember to run pytest without swallowing output flag.

py.test -s

thomas-a-neil / phission