FindIntegrationAnchors error: Error in idx[i, ] <- res[[i]][[1]]

[ksaunders73]

I'm piggy-backing off of #3930. I got the error they got, except when running FindIntegrationAnchors().

combo <- merge(sobj_tnbc, y = c(sobj_ER, sobj_her, sobj_lumb), add.cell.ids = c("TNBC", "ER", "HER","ERHER"))
head(combo@meta.data)

combo.split <- SplitObject(combo, split.by = "Patient")

combo.split <- lapply(X = combo.split, FUN = function(x) {
  x <- FindVariableFeatures(x, selection.method = "vst", nfeatures = 2000)
})

features <- SelectIntegrationFeatures(object.list = combo.split)
combo.split <- lapply(X = combo.split, FUN = function(x) {
  x <- ScaleData(x, features = features, verbose = FALSE)
  x <- RunPCA(x, features = features, npcs = 10, verbose = FALSE)
})

immune.anchors <- FindIntegrationAnchors(object.list = combo.split, anchor.features = features, 
                                         reduction = "cca", dims = 1:10)

> Error in idx[i, ] <- res[[i]][[1]] : 
  number of items to replace is not a multiple of replacement length
In addition: Warning message:
In FilterAnchors(object = object.pair, assay = assay, slot = slot,  :
  Number of anchor cells is less than k.filter. Retaining all anchors.

My dataset is very small (549 cells) and also TPM normalized, so I have a feeling that has something to do with it:

I tried setting the k.anchor and k.filter to 10 (smaller than the smallest sample of 11), and the warning goes away, but the error remains

immune.anchors <- FindIntegrationAnchors(object.list = combo.split, anchor.features = features, 
                                         reduction = "cca", dims = 1:10, k.anchor = 10, k.filter = 10)
> Error in idx[i, ] <- res[[i]][[1]] : 
  number of items to replace is not a multiple of replacement length

Thanks for reading!

[ksaunders73]

I also want to add that I tried integrating this tiny dataset with five other much larger datasets (the next smallest of the five has around 1500 cells, and the largest has around 47k cells), and I think it is definitely the size of the dataset that is the issue and not the normalization, because when I removed this tiny 549 cell dataset, I was able to integrate the other 5 together (despite one of the 5 being FPKM normalized, but being larger at around 1700 cells). When I kept the tiny 549 cell dataset in, I still get the same error as mentioned above.

[yuhanH]

hi @ksaunders73
Could you have a try that k.filter = NA and leave k.anchor to default setting?

[ksaunders73]

Hello @yuhanH! This worked to get the FindIntegrationAnchors() working, but I get the error from #3930 with IntegrateData(), and it seems lowering the k.filter method doesn't work anymore. I did since my question increase the sample size cutoff to no less than 50 cells (whereas before there were samples between 30-50 cells in them). I haven't been getting the error since making the no less than 50 cutoff, so maybe the error was occurring because some samples only had like 30 cells or 11 cells?

[yuhanH]

Hi @ksaunders73
You are right. In the IntegrateData function, we built a weight matrix between anchor cells and all other cells (k.weight).
If the object has too few cells, it may end up with an error. Also, when the object with few cells, it will violate many assumptions in our integration framework.

[ksaunders73]

Thank you very much for your help @yuhanH !

[BridgeLeeH]

Hi @ksaunders73
You are right. In the IntegrateData function, we built a weight matrix between anchor cells and all other cells (k.weight).
If the object has too few cells, it may end up with an error. Also, when the object with few cells, it will violate many assumptions in our integration framework.

Hi, a further question can be: although i know the object has too few cells, and it violates many assumptions in integration framework, then how should i address it?
I could not fix my bug when doing integrating analysis----''Error in idx[i, ] <- res[[i]][[1]] : number of items to replace is not a multiple of replacement length''. which scripts and where should i add in my current script?
Current scripts looks like:
immune.anchors <- FindIntegrationAnchors(object.list = c(E1, E2, C1, C2), anchor.features = 10000,
dims = 1:50, verbose = FALSE)
immune.integrated <- IntegrateData(anchorset = immune.anchors,dims=1:50, verbose = FALSE)

Step 1 has been finished, but step 2 runs into Error like mentioned above.

Thanks for your time.

[yuhanH]

Hi, @BridgeLeeH
Maybe you can merge the small objects into the bigger one.