Sparkline enhancements

[JobLeonard]

• [ ] panning and zooming
• [x] sort rows by row metadata
  • [x] multiple sort keys, like with columns
• [x] colour legend for "show cell attribute" choice in the settings bar
  • for example, choosing "Class" and displaying that as "categorical" shows colours, but doesn't tell us what the colours stand for right now; it's cool that most people in the research group have been able to reverse engineer, what colour refers to what by selecting a few marker genes, but that shouldn't be a required rite of passage to use the tool
• [ ] when "categorical" is chosen, colour the sparklines to match, like in linnarssonlab/cortex:
• [ ] when hovering over a graph, show cell metadata for cells under cursor?
  • [ ] determine relevant metadata. All metadata that wasn't automatically generated by Loom, perhaps?

[JobLeonard]

Working on a simple way to group sparklines, but it has some warts to work out:

[JobLeonard]

OK, we have three different but overlapping features that interact in a way that makes them very confusing at the moment:

• the legend on top of the sparklines
  • by any attribute, except those that have only unique values, or only a single value
• the ability to sort the sparklines (also extends to metadata page)
  • by attribute, except those that have only a single value
  • by gene
• the ability to group the sparklines (possible to extend to metadata page)
  • by any attribute, except those that have only unique values, or only a single value

These all have different interactions. Legend and sorting for example:

Original order:

Louvain-Jacard:

Louvain-Jacard, Cdk1:

Louvain-Jacard, Top2a:

Louvain-Jacard, Top2a, Cdk1

Louvain-Jacard, Cdk1, Top2a:

(this used to be a bigger issue when we only had heatmap plots)

Clearly grouping and sorting are related: we first group, then we sort within the groups.

Grouping and legend can become confusing if we group by one attribute, but use a different one for the legend. There may be situations where this is useful (say, group by cluster, but use tissue as the legend to see how the clustering worked out), but I think it makes more sense to tie grouping and legend to the same attribute.

Another issue is that sometimes we have a lot more groups than twenty. The OligosAll set has 266 clusters. Below is the result of gouping the first 100, and putting the rest in an "other" group:

One quick fix is to combine every group after the first twenty groups into one "other" group:

It doesn't really feel satisfactory though, especially since we currently only show the twenty biggest groups. So this needs to be reworked.