monarch plot theme

[bschilder]

I think it would be nice to have a theme_monarch function that gives plots some useful attributes.
Relates to #24.

1.

• Provide consistent colors for categories, predicates, etc. Atm the colors are just the ggplot defaults, which randomly assign the default colors to whichever variable comes first. This makes it harder to compare plots with semi-overlapping sets of categories.

For example

immune <- monarch_search("immune system",  category = "biolink:Disease")|>
	expand(predicates = "biolink:subclass_of") |>
	expand(predicates = "biolink:has_phenotype",
         categories = "biolink:PhenotypicFeature") |>
	expand(categories = "biolink:Gene")

immune|>
	activate(nodes)|> 
	filter(pcategory %in% c("biolink:Disease", "biolink:PhenotypicFeature")) |> 
	activate(edges)|>
  filter(tidygraph::edge_is_between(which(nodes(g)$pcategory=="biolink:Disease"),
  																	which(nodes(g)$pcategory=="biolink:Disease"), 
  																	ignore_dir=TRUE)) |>
	activate(nodes) |>
	filter(!node_is_isolated()) |>
	plot(layout="fr")

PhenotypicFeature is turquoise:

immune|>
	activate(nodes)|> 
	sample_n(1000) |>
	activate(edges)|>
  filter(tidygraph::edge_is_between(which(nodes(g)$pcategory=="biolink:Disease"),
  																	which(nodes(g)$pcategory=="biolink:Disease"), 
  																	ignore_dir=TRUE)) |>
	activate(nodes) |>
	filter(!node_is_isolated()) |>
	plot(layout="fr")

PhenotypicFeature is now blue:

2.

• Use different palettes for categories vs. predicates vs. etc. This avoids using the same colors to indicates different concepts.

3.

• Could even use Monarch Initiative-inspired color palette. Grabbed these from the site:

palette <- c("#2c8c9a","#79e4e7","#a7ecf3","#fafafa","white")

4.

• Could even use the same icons that Monarch uses for Phenotypes/Disease/Genes/etc. Image resources here.

5.

• Pretty layouts, especially when plotting in different common use cases (eg gene-pheno-disease relationships with lots of connections). Edge bundling with gggraph::geom_edge_bundle_force2 can be very helpful here for making hairballs more manageable.

[oneilsh]

Very much agreed and great ideas, thanks for working on this! One challenge might be that different KGs have different category/relationship labels, so hard-coding a palette against them could be tricky. One option is to use the kg prefs feature (I'm starting to wonder how much to lean on that...).

Another trick I've used in the past for similar issues is to generate a largish palette, and then hash labels and map them to an int mod the palette size to identify a psuedo-random color choice that is repeatable.

[bschilder]

Very much agreed and great ideas, thanks for working on this! One challenge might be that different KGs have different category/relationship labels, so hard-coding a palette against them could be tricky. One option is to use the kg prefs feature (I'm starting to wonder how much to lean on that...).

Very true, I was thinking of only using the hard-coded palette when the KG is the official Monarch KG.

Another trick I've used in the past for similar issues is to generate a largish palette, and then hash labels and map them to an int mod the palette size to identify a psuedo-random color choice that is repeatable.

For all other KGs, doing something like this seems like the best way to go.