Deciphering Cancer Genomes with GenomeSpy: A Grammar-Based Visualization Toolkit

  1. Kari Lavikka
  2. Jaana Oikkonen
  3. Yilin Li
  4. Taru Muranen
  5. Giulia Micoli
  6. Giovanni Marchi
  7. Alexandra Lahtinen
  8. Kaisa Huhtinen
  9. Rainer Lehtonen
  10. Sakari Hietanen
  11. Johanna Hynninen
  12. Anni Virtanen
  13. Sampsa Hautaniemi

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Background

Visualization is an indispensable facet of genomic data analysis. Despite the abundance of specialized visualization tools, there remains a distinct need for tailored solutions. However, their implementation typically requires extensive programming expertise from bioinformaticians and software developers, especially when building interactive applications. Toolkits based on visualization grammars offer a more accessible, declarative way to author new visualizations. Nevertheless, current grammar-based solutions fall short in adequately supporting the interactive analysis of large data sets with extensive sample collections, a pivotal task often encountered in cancer research.

Results

We present GenomeSpy, a grammar-based toolkit for authoring tailored, interactive visualizations for genomic data analysis. Users can implement new visualization designs with little effort by using combinatorial building blocks that are put together with a declarative language. These fully customizable visualizations can be embedded in web pages or end-user-oriented applications. The toolkit also includes a fully customizable but user-friendly application for analyzing sample collections, which may comprise genomic and clinical data. Findings can be bookmarked and shared as links that incorporate provenance information. A distinctive element of GenomeSpy’s architecture is its effective use of the graphics processing unit (GPU) in all rendering. GPU usage enables a high frame rate and smoothly animated interactions, such as navigation within a genome. We demonstrate the utility of GenomeSpy by characterizing the genomic landscape of 753 ovarian cancer samples from patients in the DECIDER clinical trial. Our results expand the understanding of the genomic architecture in ovarian cancer, particularly the diversity of chromosomal instability. We also show how GenomeSpy enabled the discovery of clinically actionable genomic aberrations.

Conclusions

GenomeSpy is a visualization toolkit applicable to a wide range of tasks pertinent to genome analysis. It offers high flexibility and exceptional performance in interactive analysis. The toolkit is open source with an MIT license, implemented in JavaScript, and available at https://genomespy.app/ .

Article activity feed

  1. GigaScience

    AbstractBackground Visualization is an indispensable facet of genomic data analysis. Despite the abundance of specialized visualization tools, there remains a distinct need for tailored solutions. However, their implementation typically requires extensive programming expertise from bioinformaticians and software developers, especially when building interactive applications. Toolkits based on visualization grammars offer a more accessible, declarative way to author new visualizations. Nevertheless, current grammar-based solutions fall short in adequately supporting the interactive analysis of large data sets with extensive sample collections, a pivotal task often encountered in cancer research.Results We present GenomeSpy, a grammar-based toolkit for authoring tailored, interactive visualizations for genomic data analysis. Users can implement new visualization designs with little effort by using combinatorial building blocks that are put together with a declarative language. These fully customizable visualizations can be embedded in web pages or end-user-oriented applications. The toolkit also includes a fully customizable but user-friendly application for analyzing sample collections, which may comprise genomic and clinical data. Findings can be bookmarked and shared as links that incorporate provenance information. A distinctive element of GenomeSpy’s architecture is its effective use of the graphics processing unit (GPU) in all rendering. GPU usage enables a high frame rate and smoothly animated interactions, such as navigation within a genome. We demonstrate the utility of GenomeSpy by characterizing the genomic landscape of 753 ovarian cancer samples from patients in the DECIDER clinical trial. Our results expand the understanding of the genomic architecture in ovarian cancer, particularly the diversity of chromosomal instability. We also show how GenomeSpy enabled the discovery of clinically actionable genomic aberrations.Conclusions GenomeSpy is a visualization toolkit applicable to a wide range of tasks pertinent to genome analysis. It offers high flexibility and exceptional performance in interactive analysis. The toolkit is open source with an MIT license, implemented in JavaScript, and available at https://genomespy.app/.

    A version of this preprint has been published in the Open Access journal GigaScience (see paper https://doi.org/10.1093/gigascience/giae040), where the paper and peer reviews are published openly under a CC-BY 4.0 license. These peer reviews were as follows:

    Reviewer 3: Luca Beltrame

    Lavikka and coworkers present an interesting visualization framework and associated application for genomics visualization. The challenges outlined by the authors in finding appropriate visualization tools for large-scale genomics data were also experienced by this reviewer, and thus better and improved tools are always welcome.

    The manuscript is well laid out, presenting the key facts in a proper manner. The use of GPU rendering for graphs is an excellent move, and I expect to be extremely useful even for machines with lower-end GPUs. The code looks reasonably written and commented (being an application, this too is important for a review). I have also tested the examples, and indeed the software is very useful (the documentation should, however, point out that some issues regarding saving the canvas still exist). One may argue that the use of JSON for the graph grammar can be awkward, but at the same time other file formats may be more problematic and/or require specialized parsers (which open yet another can of worms).

    Documentation is also logically organized. As a minor suggestion, the authors may want to add some form of search to their documentation page.

    There are is an open questions that the authors may want to answer: they explicitly mention GISTIC 1.0 for the G-score plots. Is there a specific reason why they chose 1.0? The 2.0 algorithm is far more robust and produces more reliable results.

  2. GigaScience

    AbstractBackground Visualization is an indispensable facet of genomic data analysis. Despite the abundance of specialized visualization tools, there remains a distinct need for tailored solutions. However, their implementation typically requires extensive programming expertise from bioinformaticians and software developers, especially when building interactive applications. Toolkits based on visualization grammars offer a more accessible, declarative way to author new visualizations. Nevertheless, current grammar-based solutions fall short in adequately supporting the interactive analysis of large data sets with extensive sample collections, a pivotal task often encountered in cancer research.Results We present GenomeSpy, a grammar-based toolkit for authoring tailored, interactive visualizations for genomic data analysis. Users can implement new visualization designs with little effort by using combinatorial building blocks that are put together with a declarative language. These fully customizable visualizations can be embedded in web pages or end-user-oriented applications. The toolkit also includes a fully customizable but user-friendly application for analyzing sample collections, which may comprise genomic and clinical data. Findings can be bookmarked and shared as links that incorporate provenance information. A distinctive element of GenomeSpy’s architecture is its effective use of the graphics processing unit (GPU) in all rendering. GPU usage enables a high frame rate and smoothly animated interactions, such as navigation within a genome. We demonstrate the utility of GenomeSpy by characterizing the genomic landscape of 753 ovarian cancer samples from patients in the DECIDER clinical trial. Our results expand the understanding of the genomic architecture in ovarian cancer, particularly the diversity of chromosomal instability. We also show how GenomeSpy enabled the discovery of clinically actionable genomic aberrations.Conclusions GenomeSpy is a visualization toolkit applicable to a wide range of tasks pertinent to genome analysis. It offers high flexibility and exceptional performance in interactive analysis. The toolkit is open source with an MIT license, implemented in JavaScript, and available at https://genomespy.app/.

    A version of this preprint has been published in the Open Access journal GigaScience (see paper https://doi.org/10.1093/gigascience/giae040), where the paper and peer reviews are published openly under a CC-BY 4.0 license. These peer reviews were as follows:

    **Reviewer 2: Alessandro Romanel **

    In this article, the authors introduce GenomeSpy, a grammar-based toolkit for creating customized, interactive visualizations for genomic data analysis. I find the article extremely interesting, and I believe the framework introduced by the authors has broad utility. The website is well-maintained and documented, and I particularly found the examples mentioned in the paper to be useful and informative. The authors chose to present their toolkit by narrating the navigation of a dataset generated in the DECIDER study. While the narrative makes the utility of the visualizations clear in data interpretation, what is not clear at all is how easy it is to use GenomeSpy to create those same visualizations. I believe that the success of a toolkit like this is strongly tied to its ease of use, and this aspect is not clear or prominently highlighted in the manuscript. Additionally, it would be interesting to more clearly highlight GenomeSpy's strengths compared to other approaches. By combining Rshiny and ggplot, it is indeed possible to create complex interactive data visualizations. Therefore, it would be necessary to more strongly emphasize what the other innovative aspects of GenomeSpy are, beyond GPU acceleration, compared to other approaches available today.

  3. GigaScience

    AbstractBackground Visualization is an indispensable facet of genomic data analysis. Despite the abundance of specialized visualization tools, there remains a distinct need for tailored solutions. However, their implementation typically requires extensive programming expertise from bioinformaticians and software developers, especially when building interactive applications. Toolkits based on visualization grammars offer a more accessible, declarative way to author new visualizations. Nevertheless, current grammar-based solutions fall short in adequately supporting the interactive analysis of large data sets with extensive sample collections, a pivotal task often encountered in cancer research.Results We present GenomeSpy, a grammar-based toolkit for authoring tailored, interactive visualizations for genomic data analysis. Users can implement new visualization designs with little effort by using combinatorial building blocks that are put together with a declarative language. These fully customizable visualizations can be embedded in web pages or end-user-oriented applications. The toolkit also includes a fully customizable but user-friendly application for analyzing sample collections, which may comprise genomic and clinical data. Findings can be bookmarked and shared as links that incorporate provenance information. A distinctive element of GenomeSpy’s architecture is its effective use of the graphics processing unit (GPU) in all rendering. GPU usage enables a high frame rate and smoothly animated interactions, such as navigation within a genome. We demonstrate the utility of GenomeSpy by characterizing the genomic landscape of 753 ovarian cancer samples from patients in the DECIDER clinical trial. Our results expand the understanding of the genomic architecture in ovarian cancer, particularly the diversity of chromosomal instability. We also show how GenomeSpy enabled the discovery of clinically actionable genomic aberrations.Conclusions GenomeSpy is a visualization toolkit applicable to a wide range of tasks pertinent to genome analysis. It offers high flexibility and exceptional performance in interactive analysis. The toolkit is open source with an MIT license, implemented in JavaScript, and available at https://genomespy.app/.

    A version of this preprint has been published in the Open Access journal GigaScience (see paper https://doi.org/10.1093/gigascience/giae040), where the paper and peer reviews are published openly under a CC-BY 4.0 license. These peer reviews were as follows:

    Reviewer 1: Andrea Sboner

    In this manuscript, the authors present Genome Spy, a visualization toolkit geared toward the rapid and interactive exploration of genomic features. They demonstrate how this tool can help investigators explore a large cohort of 753 ovarian cancers sequenced by whole-genome sequencing (WGS). By using the tool, they were able to identify outliers in the dataset and refine their diagnosis. The tool is inspired by Vega-lite, a high-level grammar for interactive graphics, and extends it for genomic applications.

    The manuscript is clearly written, and the authors provide links to the applications itself, tutorials and examples. I want to commend them for doing this. This is a tool that would nicely complement others and has a specific advantage of using high-performance GPUs that are now common in modern computers.

    The only concern that I have is about a couple of claims that may not be fully supported by the data provided:

    1. Claim: users can implement new visualization designs easily. While the grammar certainly enables the users to define new designs, I do not think that this is necessarily easy, as the authors themselves recognize in the discussion section when they suggest providing templates to reduce the learning curve. Indeed, the example in Figure 2 is still quite verbose and would need some time for anyone to understand the syntax and the style. The playground web application facilitates testing it, though.
    2. Claim: the grammar-based approach allows to be mixed and matched. I did not find any specific example of how to do this. It would have been quite interesting to see the intersection between the DNA representation of structural variants and RNA-seq data (if this is what it means as "mix and match").
  4. Version published to 10.1101/2023.10.06.561159v1 on bioRxiv