haohanwang / dmm

DMM (Deep Mixed Model)

Implementation of DMM in this paper:

Wang, H., Yue, T., Yang, J., Wu, W., & Xing, E. P. (2019). Deep mixed model for marginal epistasis detection and population stratification correction in genome-wide association studies. BMC bioinformatics, 20(23), 1-11.

Introduction

DMM (Deep Mixed Model) aims to extend the power of linear mixed model in correcting population stratification and identifying the associated genetic markers with the universarial approximation power of deep neural networks to model epistasis.

Replication: This repository serves for the purpose to guide others to use our tool, if you are interested in the scripts to replicate our results, please contact us and we will share the repository for replication. Contact information is at the bottom of this page.

File Structure:

• model/ main method for DMM
• libs/ other helper files, and definitions of the two components.

Usage:

Instructions

• Use loadData() in libs/helpingFun.py to load data. Users need to specify which data to load here.
• Run model.py in model/ to run the program.
• Parameter settings are in the model.py
  • parameters

    • epochs for both CNN and LSTM
    • learning rates for both CNN and LSTM
    • batch size
    • hidden state size (we use a percentage to control the hidden state size)
    • dropout rate

Data Support

• DMM supports .npy file, as shown in loadData() in libs/helpingFun.py
  • X.npy is a n x p matrix, with n samples and p SNPs
  • Y.npy is a n x 1 matrix, with n samples and 1 phenotype
• If one needs help to convert Plink data format into .npy data format, this script can be helpful.

Contact

The idea of the paper is mainly from Haohan Wang · @HaohanWang

The code in this repository is mainly written by Tianwei Yue and Jingkang Yang

dmm's Issues

Runtime (Version?) Errors

Hello,

I have been attempting to run DMM on an HPC using Tensorflow 1.15.5.
I was initially crashing on line 67 of model.py "tf.reset_default_graph()".
Then I replaced all "import tensorflow as tf" statements with "import tensorflow.compat.v1 as tf" and now I crash on line 348 "runDMM(True, modelparams)" due to an error that occurs in line 89 "model2top = CNN_x(x, y, problem, n_input, n_output = n_output,".

I could use some advice on how to fix this. Thank you!