GAM (GWAS with Advanced Machine-learning)

If you didn't want who I GWAS, you don't deserve who I GAM...

This is Pei Ting Chua Chai and Omar Halawa's CSE185 project. It implements a subset of the the "--linear" GWAS method available through plink. See the plink --linear page for more details.

Install instructions

Installation requires the pandas, numpy, cyvcf2, scipy, sklearn, matplotlib, qqman, statsmodels, art libraries to be installed. You can install these with pip install or conda install:

pip install pandas==1.5.0 numpy==1.23.3 qqman==1.0.6 scipy==1.9.3 statsmodels==0.14.0 matplotlib==3.7.1 art==5.9 scikit-learn==1.2.2 cyvcf2==0.30.22

Important Note: If the above command fails in any way, try the following two commands below which install everything via pip except for cyvcf2 which is installed via conda.

pip install pandas==1.5.0 numpy==1.23.3 qqman==1.0.6 scipy==1.9.3 statsmodels==0.14.0 matplotlib==3.7.1 art==5.9 scikit-learn==1.2.2 
conda install -c bioconda cyvcf2

Once required libraries are installed, you can install gam with the following command:

git clone https://github.com/omarhalawa3301/gam.git
cd gam
python setup.py install --user

If the install was successful, typing gam --help should show a useful message.

If you get an error saying the command was not found, then run the following (with substituting the appropriate user directory):

export PATH=$PATH:/home/<user>/.local/bin

Basic usage

The basic usage of gam is (uses the --linear option, one of the 3 available modes - see gam options):

gam -l -g genotype.vcf.gz -p phenotype.phen

To run gam on a small test example (using files in this repo) (this run should not take more than 1-2 seconds):

gam -l -g example-files/test.vcf.gz -p example-files/test.phen --covar example-files/lab3_gwas.eigenvec --maf 0.05

You should then be the output assoc.linear results file and plot png file under the example-files directory (there are references under the example-files/ref_results and example-files/ref_visuals directories)

This small test should produce the following Manhattan plot and qqplot below:

To run gam on a large test example with recommended arguments (using files in this repo) (should take ~2-3 hours for this run as it contains 900,000 SNPs and 207 samples with no multithreading implemented yet):

gam -l -g example-files/lab3_gwas.vcf.gz -p example-files/lab3_gwas.phen --dir large_test_run --out large_test --covar example-files/lab3_gwas.eigenvec --maf 0.05

Similarly to the small test run, outputs should be under the example-files directory (references also found in sub-directories for cross-checking)

This test should produce the following Manhattan plot and qqplot below:

To compare this last run to the output of plink --linear, run:

plink --linear hide-covar --covar example-files/lab3_gwas.eigenvec --allow-no-sex -g example-files/test.vcf.gz -p example-files/test.phen --maf 0.05 --out plink_run

Outputs are identical for the gam -l option run

gam options

The only required inputs to gam are

1. -g GENOTYPE_FILE, --genotype GENOTYPE_FILE as a .vcf.gz file (more file type implementation to come)
2. -p PHENOTYPE_FILE, --phenotype PHENOTYPE_FILE as a .phen file
3. Either one of the three following modes:

• -l, --linear where GWAS with a normal linear-regression model is done
• TODO: --le, --linear-ensemble where GWAS is done using an ensemble of linear regression models
• TODO: --bdt, --boosted where GWAS is done using boosted decision trees to build the model

Users may additionally specify the options below: TODO:

• -m FLOAT, --maf FLOAT: float representing minor allele frequency. If specified, the script will perform further filtration on the dataset provided by excluding SNPs whose MAF values lie below the assigned threshold.

• --covar EIGENVECFILE: name of .eigenvec file from a PCA run to use in accounting for confounding factors by covariates.

• -d DIRECTORY, --dir DIRECTORY: path to output result file(s). By default, the path is the current working directory.

• -o FILE, --out file: basename of result file(s). By default, this is the basename of the genotype file.

File format

The output file format, .assoc.linear, is the same as that of the plink --linear method. See here for more details.

Contributors

This repository was generated by Pei Ting Chua Chai and Omar Halawa, with inspiration from the CSE 185 Demo Project, GenePattern Random Forest, and many other projects.

Please submit a pull request with any corrections or suggestions.