In this work, we developed a 14-layer deep neural network, DeepGWAS, to enhance GWAS signals by leveraging GWAS summary statistics (p-value, odds ratio, minor allele frequency, linkage disequilibrium score), as well as brain related functional genomic and epigenomic information (FIRE, super FIRE, open chromatin, eQTL).
DeepGWAS is maintained by Jia Wen [jia_wen@med.unc.edu] and Gang Li [gangliuw@uw.edu].
All notable changes to this project will be documented in this file.
Our DeepGWAS is tested with R 3.6.0 with keras package. See our session info here.
- R 3.6.0
- tensorflow
- keras
GWAS summary statistics are needed for enhancement. Functional annotations are also needed.
We provide our pretrained model for users to enhance GWAS signals. User can also use their own data to train a DeepGWAS network for prediction.
R CMD BATCH --no-save --no-restore '--args input_data=enhance.Rdata model=DeepGWAS_SCZ.h5 output_file="enhance.txt"' bin/03-DeepGWAS-enhance.R
R CMD BATCH --no-save --no-restore '--args input_data=train.Rdata output_file="DeepGWAS.model.h5"' bin/02-DeepGWAS-train.R
- DeepGWAS ms
- PMID: 35396580