halkoSVD 0.2.2

Fast Python/Cython implementation of the PCAone Halko algorithm

Python implementation of Halko algorithm (PCAone)

This is an implementation of the PCAone Halko algorithm in Python/Cython for genetic data. It takes binary PLINK format (*.bed, *.bim, *.fam) as input. For simplicity, mean imputation is performed for missing data.

It is inspired by the lovely PCAone software! Have a look here.

Install and build

# Install via PyPI
pip3 install halkoSVD

# Download and install in a new Conda environment
conda env create --file environment.yml

# Download and install from GitHub directly
git clone https://github.com/Rosemeis/halkoSVD.git
cd halkoSVD
pip3 install .

# You can now run the program with the `halkoSVD` command

Quick usage

Provide halkoSVD with the file prefix of the PLINK files.

# Check help message of the program
halkoSVD -h

# Extract top 10 PCs with a mini-batch size of 8192 SNPs
halkoSVD --bfile input --threads 32 --pca 10 --batch 8192 --out halko

# Increase power iterations to 16
halkoSVD --bfile input --threads 32 --pca 10 --power 16 --out halko