Code for "Extremely sparse models of linkage disequilibrium in ancestrally diverse association studies"

This repository hosts the source code and resources to reproduce the analysis and figures in the paper "Extremely sparse models of linkage disequilibrium in ancestrally diverse association studies". The paper can be cited as follows:

Pouria Salehi Nowbandegani, Anthony Wilder Wohns, Jenna L. Ballard, Eric S. Lander, Alex Bloemendal, Benjamin M. Neale, and Luke J. O'Connor (2022) Extremely sparse models of linkage disequilibrium in ancestrally diverse association studies. Nat Genet. DOI: 10.1038/s41588-023-01487-8

Repository Contents:

• a pipeline to generate inferred tree sequences from high coverage 1000 Genomes data (in the tree_seqs directory)
• scripts to generate LDGMs from the 1000 Genomes tree sequences and subsequently infer LDGM precision matrices from LDGMs (in the inferred_ldgms directory)
• code to run analyses in the paper (in the MATLAB and simulated_tree_seqs directories)

Getting Started

First clone this repo:

git clone https://github.com/awohns/ldgm_paper.git

Downloading LDGMs inferred from the 1000 Genomes Project

The data directory can be populated with data from this Zenodo repository containing the precision matrices, LDGMs, tree sequences, correlation matrices, and genotypes used and produced by this study.

Most users will only want to download precision matrices relevant to the ancestry group(s) they are working with. These can be downloaded as follows:

make continent 

Where AFR, AMR, EAS, EUR, or SAS can be specified as "continent".

If you intend to reinfer LD precision matrices (code for doing this is in the inferred_ldgms directory), you must download the population data file:

make pops_genos_ids

To download all of the data produced in this paper (Warning! This is 43 GB) run:

cd data
make download_all

Required Software

Installing required python modules

python -m pip install -r requirements.txt

LDGM MATLAB Software

Please clone the ldgm Github repository in the home directory of this repo:

git clone https://github.com/awohns/ldgm.git

Many of the analyses require the MATLAB scripts contained in this directory (the ldgm python package is installed via the requirements.txt file).

Required software for preparing real data

Additionally, we require BCFtools, SAMtools, and HTSlib to prepare the variant data files from the 1000 Genomes Project.

Inferring Tree Sequences from Real Data

To infer tree sequences from 1000 Genomes data, navigate to the tree_seqs directory and run the following for each chromosome (replacing "chr1" with the chromosome you wish to infer):

make 1kg_chr1.trees

This command runs the following process:

1. Download NYGC high-coverage 1000 Genomes VCF data
2. Create tsinfer SampleData files from this data (excluding trios)
3. Infer a tree sequence from this data with all biallelic SNPs, mapping indels to the tree sequence with parsimony

Inferring LDGMs and LDGM precision matrices

1. After inferring tree sequences, navigate to the inferred_ldgms directory to generate LDGMs and LDGM precision matrices.

2. Run the following to generate bed files of the LD blocks (this will create one bed file per chromosome).

make EUR_LD_blocks.chr.bed

3. Next, run_make_LDGM_chr.sh contains code to create LDGMs from the inferred tree sequences. This bash script is written for the Broad Institute's internal UGER research computing cluster and is simply used to allocate resources to run make_LDGMs.py for every LD block. The script is designed to be run for each chromosome: currently chromosome 21 is specified, but any autosome can be run by modifying the value of CHR= and modifying the task list to reflect the number of blocks in that chromosome. You can modify the script to run in your preferred environment.

4. After creating LDGMs with this code, create LDGM precision matrices by running the code in run_make_precision_matrices.sh. You will need to change the commented line of code to point towards your local version of the ldgm code repository. This script is also written for the Broad Institute's internal UGER research computing cluster, but can be modified for your preferred environment: this file is simply allocating resources to run the estimatePrecision function from ldgm.

Running analyses

The MATLAB directory contains the code to run all analyses in the paper, save for the analysis in Figure 3d.

The analysis shown in Figure 3d can be run from the simulated_tree_seqs directory. Navigate to this directory and run the code in run_sim_infer_trees.sh to simulate and infer tree sequences. Then run_precision_sim.sh to infer LDGMs and LDGM precision matrices as well as calculate error metrics. As before, these bash scripts allocate resources in the Broad research computing cluster to run the computationally demanding Python and MATLAB software.

Support

Please feel free to create an issue or contact the paper's corresponding authors with any questions.