Visually Exploring the Genome At Scale

The human genome is about 2 meters long and tightly folded into the cell nucleus, a sphere that is 4 million times smaller than a pinhead. How do cells avoid entangling the DNA and ensure accessibility of necessary parts? Biologists study DNA folding through the detection of pairwise physical interactions along the DNA, which results in a 3-by-3 million pixel matrix. Visualized as a heatmap, thousands of local visual patterns become apparent. Studying these patterns is like trying to understand the average layout of parks while viewing countries on a world map. Biologists need to inspect matrix patterns and correlate them to other epigenomic profiles for sensemaking of biological features. We have developed several interactive tools to explore such large epigenomic datasets at different steps. HiGlass supports synchronized browsing and comparing of 1D and 2D data. HiPiler enables local 2D pattern exploration at scale through visual decomposition. To provide guidance during navigation and exploration in 2D maps we have developed Scalable Insets. Finally, Peax is build for interactive visual pattern search in 1D epigenomic data using unsupervised deep representation learning.