Enemble of Single-Effect Neural Networks (ESNN)

This repo is the official implementation for "Enemble of Single-Effect Neural Network" (ESNN) framework. It contains example code for running ESNN on continuous and binary classification data.

Installation and Dependencies

Python >= 3.7.4, tensorflow >= 2.1.0, tensorflow-probability >= 0.9.0, keras >= 2.3.1, matplotlib >= 3.1.2, numpy >= 1.17.2, Pillow >= 7.1.0, scikit-learn >= 0.21.3, scipy >= 1.4.1

Code

1. Regression example: ESNN_regression.py
2. Classification example: ESNN_binary.py
3. Example code to generate simulated data and to also run the "Sum of Single-Effects" regression model (SuSiE) (Wang et al. 2020): simu_example.R

To run on your own data, one can simply change the file path in the code. The simulation file contains examples of how to generate case-control data in a genome-wide association (GWA) study under the liability threshold model and a toy regression example.

RELEVANT CITATIONS

W. Cheng, S. Ramachandran, and L. Crawford. Uncertainty Quantification in Variable Selection for Genetic Fine-Mapping using Bayesian Neural Networks. iScience. 25(7): 104553.

QUESTIONS AND FEEDBACK

For questions or concerns with the ESNN software, please contact Wei Cheng or Lorin Crawford.

We welcome and appreciate any feedback you may have with our software and/or instructions.