Refit workflow with KRAS example

[khsjh]

Hello, I have some question about KRAS data processing.

1. Why reweight and reassign parameters are set to FALSE in generating slingshot object and slingshot_conditions?
   In other dataset examples, they were processed with default parameter TRUE but only KRAS example was processed with FALSE.
   Below examples are code for generating slingshot object by each dataset.

KRAS

kras <- slingshot(kras, reducedDim = "TSNE", clusterLabels = kras$Cluster,
                 start.clus = 7, extend = "n", reweight = FALSE, reassign = FALSE)
sdss <- slingshot_conditions(kras, kras$Batch, approx_points = FALSE,
                             extend = "n", reweight = FALSE, reassign = FALSE)
sdss$condition_id <- names(sdss)
sdss$mapping <- matrix(rep(1:3, each = 3), nrow = 3, ncol = 3, byrow = TRUE)
sds <- do.call(merge_sds, sdss)

Fibrosis

fibrosis <- slingshot(fibrosis, reducedDim = 'UMAP',
                 clusterLabels = colData(fibrosis)$celltype,
                 start.clus = 'AT1', approx_points = 100)

TCDD

tcdd <- slingshot(tcdd, reducedDim = "UMAP", clusterLabels = tcdd$celltype,
                  start.clus = "Central", end.clus  = "Portal")

TGFB

tgfb <- slingshot(tgfb, reducedDim = 'UMAP',
                  clusterLabels = colData(tgfb)$pheno$spatial_id,
                  start.clus = 'inner', approx_points = 100)

2. In KRAS workflow, new weights was calculated by condiments:::.sling_reassign function and it was used in evaluateK function.
   But this weights was not used in fitGAM function and just use prior slingshot object.

weights <- condiments:::.sling_reassign(sds)
icMat <- evaluateK(counts = as.matrix(assays(kras)$counts),
                   pseudotime = slingPseudotime(sds, na = FALSE),
                   cellWeights = weights,
                   conditions = factor(colData(kras)$Batch),
                   nGenes = 300,
                   k = 3:7,
                   parallel = TRUE,
                   BPPARAM = BPPARAM)

Also, below command may be not function because slingshot object was already located in SCE column and SCE recognized in column meta data at first rather than int_metadata slot.
kras@int_metadata$slingshot <- sds

kras@int_metadata$slingshot <- sds
kras <- fitGAM(counts = kras,
               conditions = factor(colData(kras)$Batch),
               parallel = TRUE,
               BPPARAM = BPPARAM,
               nknots = 6)

Slingshot::SlingshotDataSet code

#' @rdname SlingshotDataSet
#' @export
setMethod(
    f = "SlingshotDataSet",
    signature = "SingleCellExperiment",
    definition = function(data){
        if("slingshot" %in% names(colData(data))){
            return(as.SlingshotDataSet(colData(data)$slingshot))
        }
        if("slingshot" %in% names(data@int_metadata)){
            return(data@int_metadata$slingshot)            
        }
        stop('No slingshot results found.')
    }
)

Than, I can predict that command line, kras@int_metadata$slingshot <- sds was just for saving and new calculated weights was used for only evaluateK and fitGAM didn't require these weight value.
Or there are some changed logic in slingshot so that the int_metadata slot can't be used in recent version script with condiments.

I want to run my dataset by KRAS workflow but it is ambiguous to me

[HectorRDB]

Hi @khsjh
Thanks for this inquiry. A couple of comments then:

• we used different arguments for the KRAS dataset because this is what the author of the original paper also did and we wanted to reproduce their use-case. For the others we applied the recommended slingshot arguments.
• The condiments package has evolved since the paper was first written so it may not reflect the most recent version of the code, and you are correct that some parts are obselte.

Hope this helps

[khsjh]

Thank you, @HectorRDB!!
Then, I have one more question about topology test and refit process

I have multiple sample with two condition - disease vs healthy.
The topology test results showed a very low p-value, which was sufficient to reject the null hypothesis.
So, I did refit for my dataset and processed condition test. But the results of condition test were hard to interpret.
While refitting, some pseudotime values are changed and mixed. Higher pseudotime value chanhed to lower value and some NA values get pseudotime value.

I want to see the expression changes according condition specific lineages with pseudotime and cell type. But after refitting I can only identify overall difference without cell type information.

It is no matter skip refit although topology test showed almost 0 p-value (2.2e-16)?
The topology test results with KS_mean showed p.value 2.2e-16 but mmd showed 0.322.

is there any other caution point when I analyze that data?