gruyere: Genome-wide Rare Variant EnRichment Evaluation

We develop a genome-wide rare variant association test designed for identifying trait-associated loci and functional annotations. This repository accompanies our recent preprint: Leveraging functional annotations to map rare variants associated with Alzheimer’s disease with gruyere.

Installation

gruyere is written in Python. You can load gruyere along with required dependencies with the following:

git clone https://github.com/daklab/gruyere.git
cd gruyere
pip install -r requirements.txt OR conda create --name gruyere --file requirements.txt

Overview: Inputs and Outputs

Model Inputs

• G: Genotypes for N individuals and P variants [P x N]. Index should contain gene name that variant maps to. Can optionally include variant id "gene_variantID"
• Z: Functional annotations for P variants and Q annotations [P x Q]. Index should contain gene name that variant maps to. Can optionally include variant id "gene_variantID"
• XY: Individual-level covariates for N individuals and C covariates [NxC] and "Diagnosis" column for binary or continuous phenotypes

Model Outputs (Joint analysis)

• alpha.csv: Learned covariate weights by gene
• tau.csv: Learned genome-wide annotation weights
• wg.csv: Learned gene weights (mean and standard deviation)
• losses.txt: Loss per epoch
• train_performance.csv: AUC and accuracy of predictions by gene on training set
• test_performance.csv: AUC and accuracy of predictions by gene on held-out test set (optional; if using test set)

Model Outputs (Per-gene analysis)

• pvals_chr{chromosome}.csv: gene p-values and coefficients for all genes in chromosome
• preds_chr{chromosome}.csv: individual-level predictions for all genes in chromosome

Example:

• example_data/inputs.yaml contains example inputs:

---
output: '../example_outputs/' # Path where outputs are saved
XY: '../example_data/XY.csv' # File with covariates (X) and phenotypes (Y)
G: '../example_data/genotypes/' # Path to genotypes, per chromosome
Z: '../example_data/annotations/' # Path to annotations, per chromosome
epochs: 300 
n_samples: 50 # Number of times to sample the posterior to determine mean/standard deviation estimates
test_prop: 0.2 # Test set proportion
lr: 0.1
genes: '../example_data/joint_analysis_genes.txt' # List of genes to perform joint analysis on (we use FST-significant genes)
simulate: False 

Scripts

• models.py: contains gruyere model class
• data_class: processes input data and stores as dataclass object
• load_data: functions to load input data
• utils.py: utility functions
• performance.py: calculates AUROC and accuracy on gruyere predictions
• gruyere_joint.py: fits joint gruyere model
• gruyere_pergene.py: fits per-gene gruyere regression

Run gruyere

python src/gruyere_joint.py example_data/inputs.yaml
python src/gruyere_pergene.py example_data/inputs.yaml $CHR # For each chromosome