search

VEuPathDB / web-monorepo

A monorepo that contains all frontend code for VEuPathDB websites
Apache License 2.0
2 stars 0 forks source link

Bipartite Network: filter by network metrics #715

Open d-callan opened 9 months ago

d-callan commented 9 months ago

@dpbisme suggested possibly filtering nodes by degree for ex, or centrality

i assume this is by user control

asizemore commented 9 months ago

If the use case is only seeing the top or the bottom of the distribution, then i think reordering by node metric might get us to the same place but would have a simpler ux then lots of filtering. For example, any network metrics depend on the correlation and pvalue filters, becasue those affect the edges. That suggests that we put the filtering by node degree below the plot or wherever the plot controls will go (since the network might be super long, we may want plot controls elsewhere, tbd). Separating the two makes their relationship more clear, but then also we have filtering in two different places. That's why i think in the plot controls we could just have Order nodes by degree or whatever.

However, if we were to very clearly mark these plot controls as "Network Controls" and put them in this separate place, maybe it's not so bad to have a filter by degree. Still we should figure out if folks want to see the top and the bottom of the distribution or just one end still (filtering to both is more complicated, whereas ordering is easy). Another question to ask is would people want, say, the top degree and order by some other thing (phylogenetics)? Would sizing nodes by degree or centrality be acceptible?

Well i guess i have questions about this. Worth a discussion!

asizemore commented 6 months ago

Here's another thought as i'm making mockups. The bp net 2.0 viz already does a lot. I think we should make a clear boundary between doing advanced network things (anythinig harder than degree) and understanding the correlation results. Maybe that means later there's a link to "Do Hardcore Network Analysis in New App" or some different idea. But the point is i think we should keep anything more serious than ndoe degree out of the correlation bp net. One could make a nice network there and send it to a different analysis tool, but the correlation bp net viz should be about correlation results only.