LocusZoom.js: interactive and embeddable visualization of genetic association study results

1Genetic association studies are a workhorse technique in the study of complex disease genetics, and in recent years the numbers of reported studies and associations have risen dramatically. It has long been recognized that interpreting genetic association studies requires significant context, including linkage disequilibrium (LD) patterns, recombination rate, nearby genes, results for related traits, chromatin accessibility, and other information. LocusZoom is now the standard tool for visualizing GWAS results in the context of relevant genetic annotations, with both command line and web-based versions. Here, we describe a new JavaScript implementation of LocusZoom that significantly expands its capabilities, supporting interactive web-based visualizations with options for both public and private datasets.

2The basic LocusZoom style of plot has been described elsewhere. Briefly, the main area presents a scatter plot of association P-values versus chromosomal position, frequently using color to summarize LD with the lead genetic variant. Additional tracks show recombination rate, nearby genes, and relevant annotations such as variants for which associations have been reported in the EBI GWAS catalog. The region size is user selectable, typically < 1 Mb (Fig. 1).

3LocusZoom.js is an embeddable JavaScript widget for creating such visualizations, with all data rendering handled client side using web-standard technologies such as HTML, CSS and SVG. This allows significant dynamic behavior. For example, the user may hide certain elements (e.g. pseudogenes), recolor the plot based on a different LD reference population, or toggle to visualize a credible set. To facilitate exploration of a genome-wide dataset with many significant loci, the region in view may be changed either by user gestures (drag and zoom) or programmatically. At any time, the current view may be exported as PNG or high-resolution SVG files. Event callbacks allow the plot to trigger and respond to external events, such as updating a data table to reflect the information shown on the plot. Within the plot, all data elements provide tooltips that expose additional information or functionality.

4LocusZoom.js most commonly fetches data via REST APIs, and we provide an API server (https://portaldev.sph.umich.edu/docs/api/v1/) with access to standard datasets such as 1000 Genomes Project LD, GENCODE genes, HapMap-based recombination rate, and the EBI GWAS Catalog. All annotations are provided for genome builds GRCh37 and GRCh38. A plugin-based architecture makes it straightforward to add data adapters for other formats, including local and remote tabixed files. In practice, this allows plots to be generated from a static file host such as S3, reducing the barrier to entry. Since rendering is performed in the browser, the only dependency is the embedded code; private datasets do not need to be transmitted to a third-party server.

5When LocusZoom.js is integrated into existing web applications, significant customization is possible due to a track and plugin-based architecture, with a declarative layout system that can be expanded to support additional data types. LocusZoom.js has already been used to visualize PheWAS, eQTL, chromatin co-accessibility and chromatin state annotations; an example gallery demonstrates many visual and interactive features (https://statgen.github.io/locuszoom/). Track types can be layered, stacked, and dynamically added to foster exploration. LocusZoom.js has been incorporated into numerous third-party sites including the Type 2 Diabetes Knowledge Portal (https://t2d.hugeamp.org/) and PheWeb.

6Although originally developed for use in centralized data sharing portals, LocusZoom.js can be used via our web-based ‘LocalZoom’ tool (https://statgen.github.io/localzoom/) to interactively plot user-provided datasets from a local, tabix-indexed file without the need to write code. This occurs in the web browser and no data are uploaded to a central server, thus protecting privacy of sensitive results. Upon selecting the file, a modal dialog prompts the user to identify key fields such as chromosome, position, reference and alternate alleles, (-log10) P-value, effect size, standard error, and alternate allele frequency. Built-in heuristics and data validation allow LocalZoom to parse most user-provided files automatically.

7To aid exploration of results, LocalZoom exposes a number of advanced features (Supplementary Table S1). A search box allows easy access to genetic regions based on chromosome: position, gene name or rsID syntax. A 95% credible set is calculated for the region in view based on a simple Bayesian refinement method, and the visualization can be updated to reflect this calculation. Additional GWAS datasets can be added to the plot as vertically stacked panels to allow comparisons between studies, and clicking on a variant of interest will display associations from other phenotypes in a PheWAS of the UK BioBank. Data may be viewed and exported via an accompanying tabular view that includes dynamically calculated results such as credible set membership.

8Due to the size and complexity of GWAS summary statistics, some additional features require an asynchronous pre-processing step outside the web browser. We have created a new, fully interactive upload service (https://my.locuszoom.org) that allows both public and private sharing, while incorporating all of the advanced display features from LocalZoom (Supplementary Table S1). By allowing users to upload their files in advance, additional processing and annotations are possible, such as Manhattan and QQ plots, nearest gene and rsID labels, and a convenient batch view mode to browse the most significant hits across the entire genome. At present, uploaded files are limited to ∼1 GB in size. To protect sensitive data, a Google account is required to upload a study, but shared datasets can be viewed anonymously. A range of studies are provided for public browsing. In its first year of operation, this service has generated >102 000 plots across >7700 datasets. In the future, we plan to support additional features such as custom user-defined annotations and comparisons to published studies.