Challenge Prizes

  • For details on how to apply for this challenge, eligibility rules, prize breakdown, how to enter, and links to register, please visit www.Challenge.Gov.
  • For immediate inquiries, please contact
  • For information regarding the Cardiovascular Development Data Resource Center (CDDRC) see or contact jyost(at)

2022 PCGC/CDDRC Challenge Prize Winners

Shiron Drusinsky, Catherine Tcheandjieu, and Maureen Pittman

A Deep Learning Approach to Prioritize Causative Regulatory de novo Variants and Genes in Congenital Heart Disease

A recent study funded by the National Heart, Blood, and Lung Institute (NHLBI) identified 58,090 de novo variants (DNVs) in congenital heart disease (CHD) patients, but it remains unclear which DNVs are causative. Moreover, most of these DNVs are in noncoding regions of the genome, suggesting causative DNVs might contribute to CHD by affecting gene regulation. However, it is not well understood what genes are consequently dysregulated by causative regulatory DNVs, nor the mechanisms of dysregulation for any gene.

To address this problem, we developed a deep learning-based platform that takes DNA sequence as input and predicts histone marks, transcription factor binding, DNA accessibility, chromatin contact frequency, and gene expression (hereafter referred to as genomic profiles). We used this platform, which integrates the published models Akita, Sei, and Enformer, to rapidly score DNVs based on the predicted effect of the risk allele relative to the reference allele on genomic profiles in the region. As these genomic profiles amount to mechanisms of gene regulation, and because causative noncoding CHD DNVs are expected to contribute to disease by altering gene regulation, we used this platform to predict which DNVs most significantly affect gene regulation and are therefore most likely to be causative. As a proof of concept, we show that Akita-predicted effects of a CHD structural variant are experimentally reproducible.

Using our approach, we scored 58,090 CHD DNVs and found 456 DNVs whose predicted effects on gene regulation were more extreme than expected relative to rare variants and were concordant across the three models. We found that these 456 prioritized DNVs were significantly enriched near genes that have been previously implicated in CHD or are highly expressed during heart development (HEDHD), suggesting these DNVs might affect genes that play crucial roles in CHD. Using recently published snRNA-seq data, we found genes enriched with nearby prioritized DNVs exhibit differential expression between CHD subtypes and healthy hearts. Similarly, we found that genes nearby DNVs with the most extreme predicted effects are also differentially expressed between CHD subtypes and healthy hearts.

Code for the analyses presented in this Challenge solution are available at

Sheng Chih Jin, Malachi Griffith, Obi Griffith, and Amber Stratman

Role of uniparental disomy in congenital heart disease

Genetic etiologies contribute to an estimated 90% of congenital heart disease (CHD) patients, but the cause remains unknown in up to 55% of patients. Emerging evidence suggests that complex genetic explanations may be involved in CHD, but larger patient groups and more advanced analysis techniques will be needed to provide molecular diagnoses for these genetically unsolved patients. Our team at Washington University has developed TrioMix (, a framework based on Mendelian inheritance patterns that uses maximum likelihood estimation to identify uniparental disomy (UPD). Our initial analyses of 835 exome-sequencing trios in which the proband had an unselected cardiac anomaly identified two UPD cases, representing a rate of 0.24% in the initial group. The first case, diagnosed with truncus arteriosus, had a UPD(16)mat partial hetero/isodisomy – an event that was previously identified in a CHD patient with truncus arteriosus. The second case, diagnosed with a left-sided obstructive lesion, had a UPD(14)mat partial hetero/isodisomy. These findings suggest that UPD may be responsible for a small proportion of currently unexplained CHD cases.