search

msraredon / NICHES

Niche Interactions and Cellular Heterogeneity in Extracellular Signaling
https://msraredon.github.io/NICHES/
49 stars 11 forks source link

Running NICHES on large spatial datasets #37

Open RaredonLab opened 1 month ago

RaredonLab commented 1 month ago

I have encountered an interesting bug, I think related to how we are calculating edgelists in spatial data.

If I RunNICHES on imputed spatial data of reasonable size, I get the following: image and the loop quits after i=1

If I RunNICHES on the non-imputed but same spatial data, the loop gets through i=3 and then quits, and gives the following:

image

Thoughts?

I think that

  1. we probably need to update how we compute edgelists
  2. something about more vs. less sparse data is having an effect here... why? S
msraredon commented 3 weeks ago
  1. Bug #37 part A above is a data-handling bug internal to NICHES, and can be by-passed, for now, by limiting the input object to just genes in your ground truth, i.e:

fantom.features <- unique(c(NICHES::ncomms8866_mouse$Ligand.ApprovedSymbol, NICHES::ncomms8866_mouse$Receptor.ApprovedSymbol)) temp <- data.list[[i]] # copy input object to temp temp <- temp[fantom.features,] #limit to features necessary for NICHES niches.output <- RunNICHES(object = temp, ... # etc as per normal workflows

msraredon commented 3 weeks ago

Bug #37 part B above is a edge-list efficiency/scaling bug that is fixed by AOZ in the 'msbr_explorations' branch currently