MetaXcan: summary statistics based gene-level association test

MetaXcan is an extension of PrediXcan method, that infers the results of PrediXcan using only summary statistics. This repository contains a wide set of tools for calculating the gene-level association test results, and building processing pipelines as well.

A manuscript desribing the MetaXcan method and application can be found here.

The software is developed and tested in Linux and Max OS environments. The main MetaXcan script is supported in Windows.

You need Python 2.7, numpy (>=1.11.1), scipy (>=0.18.1), pandas (>=0.18.1) to run MetaXcan, and there is a GUI done in TKInter. mock is needed for the unit tests.

R with ggplot and dplyr is needed for some optional statistics and charts.

software folder contains an implementation of MetaXcan's method. The following scripts from that folder are different steps in the MetaXcan pipeline:

M00_prerequisites.py
M01_covariances_correlations.py
M02_variances.py
M03_betas.py
M04_zscores.py
MetaXcan.py

, although a typical user will use MetaXcan.py.

The rest of the scripts in software folder are python packaging support scripts, and convenience wrappers such as the GUI.

Subfolder software/metax contains the bulk of Metaxcan's logic, implemented as a python package.

MetaXcan will calculate the association results from GWAS results, as output by plink. Some support data is needed, that needs to be set up prior MetaXcan execution.

The gist of MetaXcan input is:

• A Transcriptome Prediction Model database (an example is here)
• A file with the covariance matrices of the SNPs within each gene model (such as this one)
• GWAS results (such as these, which are just randomly generated)

You can use precalculated databases, or generate new ones with tools in this repository. GTEx-based tissues and 1000 Genomes covariances precalculated data can be found here.

(Please refer to /software/Readme.md for more detailed information)

The following example assumes that you have python 2.7, numpy, and scipy installed.

1. Clone this repository.

$ git clone https://github.com/hakyimlab/MetaXcan

2. Go to the software folder.

$ cd MetaXcan/software

3. Download sample data:

# You can click on the link above or type the following at a terminal
$ wget https://s3.amazonaws.com/imlab-open/Data/MetaXcan/sample_data/v0_5/sample_data.tar.gz

This may take a few minutes depending on your connection: it has to download approximately 200Mb worth of data. Downloaded data will include an appropiate Transcriptome Model Database, a GWAS/Meta Analysis summary statistics, and SNP covariance matrices.

Extract it with:

tar -xzvpf sample_data.tar.gz

4. Run the High-Level MetaXcan Script

./MetaXcan.py \
--model_db_path data/DGN-WB_0.5.db \
--covariance data/covariance.DGN-WB_0.5.txt.gz \
--gwas_folder data/GWAS \
--gwas_file_pattern ".*gz" \
--beta_column BETA \
--pvalue_column P \
--output_file results/test.csv

This should take less than a minute on a 3GHZ computer. For the full specification of command line parameters, you can check the wiki.

The example command parameters mean:

• --model_db_path Path to tissue transriptome model
• --covariance Path to file containing covariance information. This covariance should have information related to the tissue transcriptome model.
• --gwas_folder Folder containing GWAS summary statistics data.
• --gwas_file_pattern This option allows the program to select which files from the input to use based on their name. ...This allows to ignore several support files that might be generated at your GWAS analysis, such as plink logs.
• --beta_column Tells the program the name of a column containing -phenotype beta data for each SNP- in the input GWAS files.
• --pvalue_column Tells the program the name of a column containing -PValue for each SNP- in the input GWAS files.
• --output_file Path where results will be saved to.

Its output is a CSV file that looks like:

gene,gene_name,zscore,effect_size,pvalue,var_g,pred_perf_r2,pred_perf_pval,pred_perf_qval,n_snps_used,n_snps_in_cov,n_snps_in_model
ENSG00000150938,CRIM1,4.190697619877402,0.7381499095142079,2.7809807629839122e-05,0.09833448081630237,0.13320775358,1.97496173512e-30,7.47907447189e-30,37,37,37
...

Where each row is a gene's association result:

• gene: a gene's id: as listed in the Tissue Transcriptome model. Ensemble Id for some, while some others (mainly DGN Whole Blood) provide Genquant's gene name
• gene_name: gene name as listed by the Transcriptome Model, generally extracted from Genquant
• zscore: MetaXcan's association result for the gene
• effect_size: MetaXcan's association effect size for the gene
• pvalue: P-value of the aforementioned statistic.
• pred_perf_r2: R2 of tissue model's correlation to gene's measured transcriptome (prediction performance)
• pred_perf_pval: pval of tissue model's correlation to gene's measured transcriptome (prediction performance)
• pred_perf_qval: qval of tissue model's correlation to gene's measured transcriptome (prediction performance)
• n_snps_used: number of snps from GWAS that got used in MetaXcan analysis
• n_snps_in_cov: number of snps in the covariance matrix
• n_snps_in_model: number of snps in the model
• var_g: variance of the gene expression, calculated as W' * G * W (where W is the vector of SNP weights in a gene's model, W' is its transpose, and G is the covariance matrix)

MetaXcan supports a large amount of command line parameters. Check the Github's ' wiki for those that work best for your data, and interpreting the results.

Please see the following article in the wiki.

You also have the option of installing the MetaXcan package to your python distribution. This will make the metax library available for development, and install on your system path the main MetaXcan scripts.

You can install it from the software folder with:

# ordinary install
$ python setup.py install

Alternatively, if you are going to modify the sources, the following may be more convenient:

# developer mode instalation
python setup.py develop

PIP support coming soon-ish.

Issues and questions can be raised at this repository's issue tracker.

There is also a Google Group mail list for general discussion, feature requests, etc. Join if you want to be notified of new releases, feature sets and important news concerning this software.

You can check here for release history.

Transcriptome Models are a key component of MetaXcan (and PrediXcan) input. As models are improved, sometimes the format of this databases need be changed too. We only provide support for the very latest databases; if a user updates their repository to the latest version and MetaXcan complains about the transcriptome weight dbs, please check if new databases have been published here.

For the time being, the only way to use old transcriptome models is to use older versions of MetaXcan.

Check this if you want to learn more about more general or advanced usages of MetaXcan.

Check out the Wiki for exhaustive usage information.

The code lies at

/software