De novo variants (DNVs) with deleterious effects have proved informative
in identifying risk genes for early-onset diseases such as congenital heart
disease (CHD). A number of statistical methods have been proposed for
family-based studies or case/control studies to identify risk genes by
screening genes with more DNVs than expected by chance in Whole Exome
Sequencing (WES) studies. However, the statistical power is still limited
for cohorts with thousands of subjects. Under the hypothesis that connected
genes in protein-protein interaction (PPI) networks are more likely to
share similar disease association status, we developed a Markov Random
Field model that can leverage information from publicly available PPI
databases to increase power in identifying risk genes. We identified 46
candidate genes with at least 1 DNV in the CHD study cohort, including 18
known human CHD genes and 35 highly expressed genes in mouse developing
heart. Our results may shed new insight on the shared protein functionality
among risk genes for CHD.