Chromatin Insulator Function and Nuclear Organization

National Institute Of Diabetes And Digestive And Kidney Diseases

Investigators

Linked publications, trials & patents

Abstract

Deciphering the contribution of cis-regulatory elements to gene expression using 3D interactions and graph neural networks We present a novel deep learning architecture, CreGNN, based on a graph transformer network that incorporates data from transcription factor binding, histone modifications, chromatin accessibility, and 3D physical contacts. A promoter-centered graph is built for each gene, with candidate cis-regulatory elements (CREs) filtered by 3D contacts. Our approach can ascertain key chromatin-associated factors and CREs, while quantifying their contribution to the regulation of each gene. Compared to previous methods, CreGNN exhibits higher resolution and accuracy in terms of classification and prediction. Thus far, we have analyzed publicly available data for two human cell lines, mouse embryonic stem cells, and Drosophila embryos. We validated our predictions using experimental data including CRISPR perturbations and massively parallel reporter assays. Use of 3D information and contact frequency moderately improved gene expression prediction overall, consistent with 3D interactions fine tuning expression of a subset of genes. We aspire to apply our method to additional datasets including organisms with fewer existing CRE annotations as well as develop a searchable tool for specific genes of interest. In summary, CreGNN represents a high-resolution, robust computational framework that contributes to our understanding of specific gene regulatory mechanisms at the 3D level and allows the design of targeted strategies to modify gene expression within cells. Characterization of a potential mouse model for pancreatic acinar cell carcinoma We have obtained Cela1-CreERT (tg/+);ID3 fl/fl, APC fl/fl mice induced with tamoxifen (and matched controls treated with sesame oil) generated at the National Cancer Institute (NCI) Center for Preclinical Research. Their ages range from 4 to 10 months, and we have begun to observe sick phenotypes (hunched or extended weight changes) in mice ~5-8 months old. We are monitoring and weighing the mice weekly and performing tissue and blood collections on sick mice. Our preliminary results show that ID3 fl/fl, APC fl/fl mice have disrupted pancreatic exocrine function. Initial histopathology observations include enlarged acinar cells, with corresponding enlarged nuclei. We additionally observe immune cell infiltration near pancreatic islets, which are essential for insulin production. As initial molecular characterization of these aberrant tissues, we plan to perform ATAC-seq chromatin accessibility assays and RNA-seq on whole pancreas and FACS-sorted acinar cells. This work should lead to greater insight into the development and function of an organ that can give rise to not only terminal cancer but also chronic disease such as diabetes.