WashU Epigenome Browser update 2019

Abstract

The WashU Epigenome Browser (https://epigenomegateway.wustl.edu/) provides visualization, integration and analysis tools for epigenomic datasets. Since 2010, it has provided the scientific community with data from large consortia including the Roadmap Epigenomics and the ENCODE projects. Recently, we refactored the codebase, redesigned the user interface, and developed various novel features. New features include: (i) visualization using virtual reality (VR), which has implications in biology education and the study of 3D chromatin structure; (ii) expanded public data hubs, including data from the 4DN, ENCODE, Roadmap Epigenomics, TaRGET, IHEC and TCGA consortia; (iii) a more responsive user interface; (iv) a history of interactions, which enables undo and redo; (v) a feature we call Live Browsing, which allows multiple users to collaborate remotely on the same session; (vi) the ability to visualize local tracks and data hubs. Amazon Web Services also hosts the redesign at https://epigenomegateway.org/.

Figure 1.

Component-based design of the new software architecture. The area with a darker background (A and B) represents the old design; the rest (C–J) represents the new design. (A) In the old design, a centralized server retrieves data from a remote location, processes it and sends it to the user. This creates a bottleneck, as data processing for all users must happen on the server. (B) Our old codebase lacked encapsulation of functions and organization, making it extremely difficult to maintain. (C) In the new design, clients directly request data from where it is stored, eliminating the bottleneck that (A) describes. (D) Data are fetched in parallel, boosting performance. (E) Data processing, which used to happen on the central server, instead runs on worker threads on the client’s machine. This adds parallelization and improves the software’s responsiveness. (F) Each step of the data pipeline is a separate module in our codebase, which improves organization and reduces bugs. (G) Track components or visualizers are also separated into modules. (H) A global data store orchestrates information sharing between components and powers features like Undo/Redo and history. (I) Local session storage mirrors the global data store and ensures that no work is lost if there is a crash or if the user reloads the page. (J) When a Live session is established, session data are synced with Firebase. When any device updates Firebase, the changes are pushed to the global data store (H), which updates the visualization components.

Figure 2.

A sample of the new features. (A) Serverless design. Clients directly request data from where it is stored, eliminating the bottleneck of a server. (B) Local folders or files can now be organized into a local data hub. Local files help protect data privacy and load faster than files hosted on remote servers. (C) Live Browsing can synchronize the same view across multiple computers, simplifying collaboration. (D) Virtual Reality prototype. The current prototype can visualize chromatin interactions and numerical tracks in 3D space for viewing with a VR headset.

Figure 3.

Guide to the new UI. This screenshot illustrates default tracks loaded from the Roadmap GEO public hub after adjusting track order and heights to display heat maps. (A) UI elements are modular components in the codebase; the left column indicates the names of these components. (B) Functions for each menu item. (C) The right column contains the metadata management interface and metadata colormap. Each unique color represents a metadata value, such as pink meaning IMR90. The metadata column is customizable and provides a way to select adjacent tracks that share the same metadata values. (D) The main visualization displays the HOXA gene cluster. From top to bottom: a genome ruler track, histone modification tracks, a chromatin state (chromHMM) track, a matplot track, whole genome bisulfite sequencing data in a ‘methyl’ track, a Repeatmasker track and a Gencode version 29 gene track. (E) From left to right: four toggleable tools that change how the mouse behaves, which include dragging, zooming and reordering tracks; a set of pan and zoom buttons; and the Undo, Redo and History tools.

Figure 4.

Statistics of tracks hosted by the WashU Epigenome Browser. (A) Statistics on the number of human and mouse tracks. (B) Tracks grouped by consortia. (C) Hosting location of tracks. More than half of the tracks are hosted on the cloud, with each consortium’s data portal providing access. We host the remainder on our own servers (‘on-premises’). (D) Track statistics for three genome assemblies, grouped by consortia. Tracks are categorized by experiment type, including expression, methylation, ChIP-seq & open chromatin and chromatin interaction.

Figure 5.

Cell type-specific analysis and visualization of JunD binding on transposable elements (TEs), demonstrated in the new Browser. (A) A JunD-binding peak on an MLT1 transposable element, highlighted in green. methylCRF tracks show the predicted methylation level of CpG sites, where 0 means fully unmethylated, and 1 means fully methylated. (B) A region set view of 20 TEs shows the binding happens specifically on TE elements in the K562 sample. MLT1 is the rightmost region. (C) Scatter plot shows the relationship between mean signals for K562 H3K27ac (y-axis) and K562 JunD ChIP-seq (x-axis) in the 20 TEs. Each dot represents a TE and its flanking region. (D and E) Box plots of GM12878 (D) and K562 (E) samples indicate the average signal from 20 TEs and the regions flanking them. Each region is split into 50 bins (hence the x-axis ranges from 0 to 50), and is displayed from 5′ to 3′.

Figure 6.

Chromatin domain analysis and visualization of GM12878 cells, demonstrated in the new Browser. The tracks from top to bottom are a genome ruler, GencodeV29 gene annotations, and CTCF ChIP-seq, H3K4me3, ChIA-PET and HiC data from GM12878 cells. Light blue highlighted regions indicate CTCF peaks, correlated with ChIA-PET anchors and HiC domain boundaries.