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 the 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 pre-trained model for users to enhance GWAS signals. Users 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