Dimensionality reduction methods to large-scale genotype data.

This is a repository for a source code to replicate Sakaue S et al. Nature Communications. 2020.

Overview

To find finest-scale population substructures, we applied a series of dimensionality reduction methods to large-scale genotype data of Japanese and also worldwide populations. We will introduce the python-based source code for replicating our results.

Publication

Our paper is out!

Sakaue S et al. Dimensionality reduction reveals fine-scale structure in the Japanese population with consequences for polygenic risk prediction. Nature Communications. 2020

Requirements

• umap-learn
• scikit-learn
• scipy
• pandas
• MulticoreTSNE

We would recommend you to install these dependencies by using anaconda.

In addition, plink version 1.9 (https://www.cog-genomics.org/plink2/) should be in your $PATH.

Installation

If you would like to download the scripts,

$ git clone https://github.com/saorisakaue/Genotype-dimensionality-reduction
$ cd ./Genotype-dimensionality-reduction

Input

User input is genotype data in plink format.

• plinkfile_prefix.{bed/bim/fam}

Important! To run dimensionality reduction methods effectively, we would recommend to impute all missing genotypes in your plink file. Such imputation tools are;

• Beagle (https://faculty.washington.edu/browning/beagle/beagle.html)
• Eagle (https://data.broadinstitute.org/alkesgroup/Eagle/)

Eagle is scalable and accurate for imputing biobank-scale genotype data (e.g., N > 100K).

Usage

Two scripts

• 01_prep_genotype.sh : for LD pruning and formatting of the genotype
• 02_run_dr.py : main dimensionality reduction functions

are used for performing dimensionality reduction methods on genotype data provided by the user as plink file. To run the script on the command line, simply

$ bash 01_prep_genotype.sh ${path/to/plinkfile_prefix}
$ python 02_run_dr.py ${path/to/plinkfile_prefix}

can do all the analyses.

Example genotype data of 50 individuals from 1KGPp3v5 is provided in ./data_umap directory. If you use example dataset, the command line will be as follows;

$ bash 01_prep_genotype.sh ./data_umap/1KG.selected
$ python 02_run_dr.py ./data_umap/1KG.selected

Please note that it would be efficient to split the process within 02_run_dr.py into each part of the five dimensionality reduction methods, particularly when the genotype data is large (N > 10K).

Contact

Any questions? Saori Sakaue (ssakaue[at]sg.med.osaka-u.ac.jp)