Sparse Convolutional Denoising Autoencoders for Genotype Imputation

Genotype imputation, where missing genotypes can be computationally imputed, is an essential tool in genomic analysis ranging from genome wide associations to phenotype prediction. To explore the performance of deep learning for genotype imputation, we proposed a deep model called a Sparse Convolutional Denoising Autoencoder (SCDA) to impute missing genotypes. We constructed the SCDA model by using a convolutional layer that can extract various correlation or linkage patterns in the genotype data and applying a sparse weight matrix resulted from the L1 regularization to handle high dimensional data. This study thus points to another novel application of deep learning models for missing data imputation in genomic studies.

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes. The code is performed in a Jupyter Notebook, because Jupyter Notebook provides an interactivly develop and test manner, which is very convient for practical purpose.

Prerequisites

Python3.6 
virtualenv >= 16.4.3

Setup

Create virtual environment

git clone https://github.com/shilab/SCDA.git
cd SCDA
mkdir venv
python3 -m venv venv/
source venv/bin/activate

Install requirment dependents

pip install tensorflow sklearn pandas jupyter matplotlib

Launch Jupyter

Launch Jupyter notebook by

jupyter notebook 

which will launch a browser. If you have any problem, please reference Jupyter notebook offical document

Case study

To demonstrate the SCDA's usage, we provide a case study for yeast genotype imputation, which achieves above 99% imputation accuracy.

Input file format

The input file should be CSV format with 'tab' delimiter. The header is SNPs' position, and the first column is sample ID, as following table shown. This format can be easily generated from VCF file by removing the first 8 lines and transferring the row and column.

SAMID	chrI_33070	chrI_33147	chrI_33152	chrI_33200
01_01	1	2	0	1
01_02	1	1	2	1
01_03	2	0	1	1
01_04	1	2	0	2
01_06	0	1	2	0

where 1s and 2s reprensent the variants from laboratory strain (BY) and an isolate from a vineyard (RM), respectively. 0s reprensent the mssing SNPs. Different species may have different encoding for their variants. Make sure the encodeing starts with 1, and 0s represent missing values.

Output file format

The predict results contains the probilities of each label 1, 2 for each sample, as the following table shown.

1	2
0.999944	5.559894e-5
1.9059176e-05	0.9999809
0.9999864	1.3582917e-05
0.00040740182	0.9995926
0.99998647	1.354419e-05

For the training process

1. Unzip the yeast_genotype_train.txt.zip file in data folder where you launched Jupyter Notebook.
2. Click SCDA_train.ipynb in brower. Run the cells one by one, or run the whole notebook in once. It will load training data, train a SCDA model and save the trained model.

For the imputation process

1. Unzip the yeast_genotype_test.txt.zip file in data folder where you launched Jupyter Notebook.
2. Click SCDA_test.ipynb in brower. Run the cells one by one, or run the whole notebook in once. It will load training data, load a trained SCDA model and predict the average imputation accuracy.

Citation

Junjie Chen, Xinghua Shi*. Sparse Convolutional Denoising Autoencoders for Genotype Imputation. Genes, 2019.