add support for a genomic "relative region" plot

[ngehlenborg]

User story Epigenomic signals around a particular feature type (transcription start site, location of a particular motif) which has many instances across the genome can give insight into regulation of nearby genes.

(from https://www.biorxiv.org/content/10.1101/2020.11.09.373613v1.full.pdf)

Preferred solution Vega-lite-based plot that pulls cluster profiles from a Zarr store (that is also used for genome-wide views)? The issue here is that one would need the locations of all instances of the feature and load those regions and average them (in the example above each lines is the average across all sites in a given cluster), which might be slow without a server. With a HiGlass server available, we could build this functionality directly into the server (as an extension of the work that @keller-mark did for retrieval of cluster/group summaries). Alternatively, one could create a new data type that represents such averaged profiles, which would be precomputed and can be linked to cell sets.

Possible alternatives Implementation of this plot type in Gosling for the front end? (would still require the server-side support)

[keller-mark]

The above visualization feels like a summary version of the "tornado plot". Do you think this would be useful to implement as well?

https://elifesciences.org/articles/35989/figures#fig3

https://deeptools.readthedocs.io/en/develop/content/example_gallery.html#dnase-accessibility-at-enhancers-in-murine-es-cells

[ngehlenborg]

Yes! Those are indeed closely related in terms of the data that is being shown.

[ngehlenborg]

I didn't realize that they are called tornado plots. Pretty fitting!