jr-leary7 / scissors Goto Github PK

• parallel processing sometimes errors out due to unused workers, memory limits, etc.

• Use doParallel and foreach

• Just add a CITATION file to be used à la citation("SCISSORS")

• Do some benchmarking of the various cosine distance functions available in R, & maybe add a package dependency if the improvement is significant enough over the current base R implementation

I have been attempting to run ReclusterCells on my Seurat object with the following command:

t_reclust <- ReclusterCells(data, which.clust = fibroblast_labels, merge.clusters = TRUE, is.integrated = TRUE, integration.ident = "dataset", use.sct = TRUE, k.vals = c(30, 40, 50), resolution.vals = c(.2, .3, .4), n.HVG = 4000, n.PC = 15, use.parallel = FALSE, redo.embedding = FALSE, random.seed = 312)

These is the latest print statements from the execution:

Finished calculating residuals for counts
Set default assay to SCT
|++++++++++++++++++++++++++++++++++++++++++++++++++| 100% elapsed=15m 04s
[1] 1
[1] 2
Warning: Assay integrated changing from Assay to SCTAssay
Error in { :
task 1 failed - "long vectors (argument 1) are not supported in .C"
In addition: There were 50 or more warnings (use warnings() to see the first 50)

When checking warnings(), these are the responses (the other 48 are identical to 30):

29: In glm.nb(formula = as.formula(new_formula), data = data) :
alternation limit reached
30: In theta.ml(Y, mu, sum(w), w, limit = control$maxit, trace = control$trace > ... :
iteration limit reached

Need to be able to support non-Seurat integration, and / or usage of integrated data without re-integrating each subset as that's not always desired

Use testthat package to add some basic unit tests

• Make the main function easier to use
• Also need to make sure re-integration functionality works correctly, it can get tricky when doing non-Seurat integration routines

Don't want to limit to just reclustering the seurat_clusters Ident.

• Requires actually passing R CMD CHECK and having codecov activated for the repo

Hi,

I am testing your package with my data. I have an integrated Seurat object. I used STACAS for integration and the clustering at this point is already defined as cell types strings.
I converted the strings describing cell types to integers, as it seems required by the ReclusterCells function and created a seurat_clusters variable. Clusters IDs are integers from 1 to 15.
I then run directly
t_reclust <- ReclusterCells(scdata.combined,auto=T,redo.embedding=T,n.cores=20,is.integrated = T, integration.ident="SampleName")
but I get
Error: No cells found

I could trace this error to
if (!merge.clusters) {
temp_obj <- subset(seurat.object, subset = seurat_clusters ==
which.clust[[i]])
}

The issue seems to be in
which.clust <- as.integer(names(which(ComputeSilhouetteScores(seurat.object) <
cutoff.score)))

this results in a which.clust with cluster IDs shifted by 1, i.e. if seurat_clusters range is [1,15], the which.clust range is [0,14] and as a consequence
temp_obj <- subset(seurat.object, subset = seurat_clusters == which.clust[[i]])
fails.

I am probably doing something wrong. Can you please help ?

Thanks

• Convert README.md to an RMarkdown document
• Use GitHub Actions to render upon push to master branch

Make sure the function doesn't assume that the user has employed SCTransform normalization

Want to allow for integrated dataset generated by the this workflow.

• else ... if logic should just be if ... if instead