Meta-Regression Visualization Package for Rare Variant Association Studies

Overview

This Python package provides tools for computing and visualizing the proportion of variance explained by different moderators in a meta-regression model. The package is designed for Rare Variant Association Studies (RVAS), incorporating genetic variant effect sizes, moderator variables, and statistical model fitting.

Features

• Compute weighted sum of squares (SST, SSR_full, SSR_reduced).
• Calculate proportion of variance explained for each moderator.
• Generate bar plots for variance contributions of moderators.

Installation

You can install the package locally with:

pip install -e .

or

pip install path/umr-visualizer

Usage

1. Importing the Package

import umr_visualizer as uv

2. Example Input Table

Your dataset should be structured as follows:

Gene	Variant ID	Beta	SE Beta	Mod_1	Mod_2	Mod_3	...
Gene1	G1	0.2	0.1	1.2	0.5	1.0	...
Gene1	G2	-0.5	0.2	0.8	0.3	0.7	...
Gene2	G3	1.3	0.15	1.1	0.6	0.8	...
Gene2	G4	0.7	0.05	0.9	0.4	0.9	...

Application : The dataset is now filtered specifically for the ABOP gene across all files from the directory: /scratch/groups/mrivas/larissaredo/prepped_files . Additional details on the dataset size and preprocessing could be found on from test_package.ipynb

3. Compute Proportions

# Define moderators
moderators = ["Mod_1", "Mod_2", "Mod_3"]

Compute proportions of variance explained example per moderator for a collection of genes

prop_red = uv.proportion_variance_explained_per_gene(df, fitted_values_full,"Mod1", ['Gene1', 'Gene2'])

4. Plot Proportions explained for a list genes

uv.plot_proportion_variance_explained(df, fitted_values_full,"matplotlib" , ['Gene1', 'Gene2'])

uv.plot_proportion_variance_explained(df, fitted_values_full,"seaborn" , ['Gene1', 'Gene2'])

uv.plot_proportion_variance_explained(df, fitted_values_full,"plotly" , ['Gene1', 'Gene2'])

Methods & Equations

The package follows the meta-regression framework:

1. Weighted Mean

   $$\bar{y}_w = \frac{\sum w_i y_i}{\sum w_i}$$

   where $$w_i = \frac{1}{SE_{\beta_i}^2}$$.

2. Total Sum of Squares (SST)

   $$SST = \sum w_i (y_i - \bar{y}_w)^2 $$

3. Sum of Squared Residuals (SSR_full & SSR_reduced)

   $$SSR_{\text{full}} = \sum w_i (y_i - \hat{y}_i)^2 $$

   $$SSR_{\text{reduced}, j} = \sum w_i (y_i - \hat{y}_i^{\text{reduced}})^2 $$

4. Proportion of Variance Explained

   $$Proportion_j = \frac{R^2_j}{R^2_{\text{total}}} $$

   with

   $$R^2_j = \frac{SSR_{\text{reduced}, j} - SSR_{\text{full}}}{SSR_{\text{reduced}, j}} $$

Directory Structure

meta_regression_viz/

│
├── images/
│   ├── output.png         # Image examples 
├── meta_regression_viz/
│   ├── __init__.py         # Package initializer
│   ├── stats.py            # Functions for computing SST, SSR, and proportions
│   ├── plots.py            # Functions for plotting variance explained
│
├── setup.py                # Installation script
├── README.md               # Documentation
├── LICENSE                 # License file
├── requirements.txt        # List of dependencies
├── test_package.ipynb       # Example Update
└── .gitignore              # Ignore unnecessary files

License

This package is licensed under the MIT License.

Authors

• Salma Zainana
• szainana@stanford.edu +++