Improve readability of boxplots that have too many boxes

[logstar]

Boxplots transferred by /tpm/gene-disease-gtex/plot and /tpm/gene-all-cancer/plot are 13 inches wide and 9 inches long. Following is an example:

When there are a couple of more boxes, the boxplot will become less readable, because smaller boxes and texts will be shown in the plot.

@taylordm suggested to collapse all GTEx tissue groups into one box. Implement this approach for the new endpoints in #44. The plots will be further discussed and evaluated.

cc @chinwallaa @afarrel @taylordm

[logstar]

All-cancer-groups and all-GTEx-tissue-subgroups plot is not easily readable without collapsing GTEx samples. For example:

Collapsing GTEx samples into one box makes the plot easily readable. For example:

I will implement #44 with all GTEx samples collapsed into one box.

[logstar]

@taylordm suggested to provide another API option/parameter to return wider images for boxplots with too many boxes. For example, add a gtex_box_group parameter, which takes the following values:

• tissue_subgroup: plot all GTEx tissue subgroups. Return a wider image.
• collapse: Collapse all GTEx samples into one box. Return a narrower image.

Wider images will require horizontal scrolling on PedOT website, so this needs to be further coordinated with FNL team front-end development.

The width of the image can be implemented either statically or dynamically. I will implement the static-image-size option, to make the API more robust and straightforward.

• Static image size:
  • Pros: Natively supported by plumber. Straightforward implementations.
  • Cons: Two image sizes require two endpoints, as described in https://www.rplumber.io/articles/rendering-output.html#customizing-image-serializers on 10/18/2021. The image size of each type of plot also needs to be manually selected for every new database version, in order to accommodate more or less boxes. This implementation will introduce code duplications.
• Dynamic image size:
  • Pros: Two image sizes only require one endpoint. The image size will be dynamically set for different number of boxes in different database versions. This implementation will not have duplicated code.
  • Cons: Not natively supported by plumber. I have created an issue in plumber to request this feature, https://github.com/rstudio/plumber/issues/837. Dynamically selected image sizes may cause errors, such that the image will not be available.

cc @taylordm @afarrel @chinwallaa

[logstar]

@taylordm @afarrel @chinwallaa - Following is an example all-cancer all-GTEx-tissue boxplot with image width doubled. The response time of the wide image is about 80% longer than the standard image, because the wide and standard image sizes are about 600KB and 1200KB respectively.