Genetic and Epigenetic Gene Regulation

Motivated by the fact that most disease associated variants identified to date are located in non-coding parts of the genome, which likely harbors regulatory elements, we are studying the effect of naturally occurring sequence variation on gene regulation. To characterize regulatory sequence variants two related challenges have to be met: 1) regulatory elements have to be recognized and 2) the corresponding target genes have to be identified. Epigenetic marks such as histone modifications have proved instrumental for the identification of regulatory elements in the genome, while the integrated analysis of genetic variation and gene expression provides a strategy (expression QTL mapping) to identify targets of regulatory variants. Ultimately the integration of genetic, genomic and epigenomic data set is expected to lead to a comprehensive understanding of regulatory sequence variation and its role in disease. Towards these goals we have:

• developed a random walk approach to identify the regulatory networks underlying common disease from DNA-methylation data (Hawe Nature Genetics 2022) - code
• extended this approach using Bayesian graphical models to make full use of prior networks and the correlation structure in the data (Hawe Genome Medicine 2022) - code
• integrated proteomics and gene expression to identify cis and trans acting eQTL/pQTL in the human atrial appendage (Assum Nature Communications 2022) - code
• reviewed current methods for network reconstruction from multi-omics data (Hawe Frontiers in Genetics 2019)
• performed one of the largest eQTL studies to date in the human heart (Heinig Genome Biology 2017)
• developed the computational tool histoneHMM for the identification of differentially modified regions for histone modifications with broad genomic footprints (Heinig BMC Bioinformatics 2015). developed predictive models to identify functional genomic elements predictive of regulatory variants (Budach, Heinig* and Marsico* Genetics 2016)
• performed an integrated analysis of the consequences of genetic variation for multiple levels of epigenetic and transcriptional regulation (Rintisch*, Heinig* Genome Res 2014)
• developed a statistical approach for the identification of a transcription factor driven regulatory network, including its master regulator and the interpretation of disease association (type 1 diabetes) using this regulatory network (Heinig Nature 2010)
• developed the computational tool sTRAP for the identification of causative cis regulatory variants affecting transcription factor binding (Manke*, Heinig* Hum Mutat 2010) and successfully applied this tool in a disease gene study (Monti Nat Genet 2008) for heart failure

Complex traits are associated with houndreds if not thousands of non-coding variants throughout the whole genome. Theoretical models such as the omnigenic core gene model have been proposed to reconcile this observed genetic architecture with the potential molecular mechanisms: when small effects of multiple risk loci converge on the same downstream core genes in regulatory networks, a large proportion of the heritability can be explained. The key challenges are that downstream targets are difficult to identify using QTL data and that the core genes for specific diseases are unknown.

To adress these challenges, we have:

• developed Speos - a graph neural network approach to predict core genes of complex disease from network, GWAS and gene expression data (Ratajzcak bioRxiv 2023) - code - docs
• developed a data integration approach that makes use of polygenic risk scores and pathway annotations to identify trans-acting QTL from protein and transcript expression data. We applied it to the atrial fibrillation cohort of the symAtrial consortium to identify candidate core genes of atrial fibrillation (Assum Nature Communications 2022) - code