Abstract
Extensive efforts have been made to understand genomic function through both experimental and computational approaches, yet proper annotation still remains challenging, especially in non-coding regions. In this manuscript, we introduce GenoSkyline, an unsupervised learning framework to predict tissue-specific functional regions through integrating high-throughput epigenetic annotations. GenoSkyline successfully identified a variety of non-coding regulatory machinery including enhancers, regulatory miRNA, and hypomethylated transposable elements in extensive case studies. Integrative analysis of GenoSkyline annotations and results from genome-wide association studies (GWAS) led to novel biological insights on the etiologies of a number of human complex traits. We also explored using tissue-specific functional annotations to prioritize GWAS signals and predict relevant tissue types for each risk locus. Brain and blood-specific annotations led to better prioritization performance for schizophrenia than standard GWAS p-values and non-tissue-specific annotations. As for coronary artery disease, heart-specific functional regions was highly enriched of GWAS signals, but previously identified risk loci were found to be most functional in other tissues, suggesting a substantial proportion of still undetected heart-related loci. In summary, GenoSkyline annotations can guide genetic studies at multiple resolutions and provide valuable insights in understanding complex diseases. GenoSkyline is available at http://genocanyon.med.yale.edu/GenoSkyline. Genome-wide association study has been a productive approach to studying human complex diseases in the past decade, yet challenges still remain in both identifying and interpreting disease-associated loci. In our previous work, we showed that integrated analysis of GWAS summary statistics and genomic functional annotations could enhance the performance of signal prioritization. In this paper, we further improve our annotation through integrating a rich collection of epigenomic data from the Roadmap Epigenomics Project. We introduce GenoSkyline, a statistical framework to predict tissue-specific functional regions in the human genome, and demonstrate its ability to capture tissue-specific functionality through extensive case studies. We then illustrate a variety of ways that GenoSkyline could benefit post-GWAS analysis. The performance of signal prioritization is further improved using annotations of disease-related tissue types. Furthermore, combining GenoSkyline annotations with GWAS results allow us to partition heritability by tissue types and generate new hypotheses regarding the disease etiology behind each risk locus, thereby providing novel biological insights to many human complex diseases. GenoSkyline is powerful, robust, and customizable. We believe that GenoSkyline and its applications can guide genetics research at multiple resolutions and greatly benefit the broader scientific community.
All Related Versions

This publication has 41 references indexed in Scilit: