Confusing steps in tutorial of recursive MuSiC algorithm?

[tjbutler003]

Hello Xuran, Thank you very much for creating this package, we have found it helpful and are still trying to figure out some issues. I have been trying to follow the recursive algorithm explanation: [https://xuranw.github.io/MuSiC/articles/MuSiC.html#estimation-of-cell-type-proportions-with-pre-grouping-of-cell-types]

I have been having trouble with the IEmarkers object, when I downloaded the IEmarkers.RData object, I found an "Immune.marker" and "Epith.marker" object. I assumed I needed to turn them into a list, which I did with IEmarkers = list(C3 = Epith.marker, C4 = Immune.marker) However I get an error:

Error in music_prop.cluster(bulk.eset = mouse.bulk, sc.eset = mouse.sc, : Cluster number is not matching!

when typing in the command you provided. I also see that in the tutorial, the command: Est.mouse.bulk = music_prop.cluster(bulk.eset = Mouse.bulk.eset, sc.eset = Mousesub.eset, group.markers = IEmarkers, clusters = 'cellType', group = 'clusterType', samples = 'sampleID', clusters.type = clusters.type) uses 'group', but the options from the Documentation have the name of the argument as 'groups', is that the one we should be using? Any help is appreciated.

Thanks, T.J.

[Prayforhanluo]

same issue.

[tkamth]

Same issue here, but I have tried the following, it would be good if @xuranw can confirm this.

Neutro<-NULL Podo<-NULL Endo<-Epith.marker Macro<-Immune.marker IEmarkers<-list("C1"=Neutro,"C2"=Podo,"C3"=Endo,"C4"=Macro) Est.mouse.bulk = music_prop.cluster(bulk.eset = Mouse.bulk.eset, sc.eset = Mousesub.eset, group.markers = IEmarkers, clusters = 'cellType', group = 'clusterType', samples = 'sampleID', clusters.type = clusters.type)

[solinvicta]

I tried the steps that @tkamth had proposed, and I was able to run it without error; however, the cell proportions seemed a little odd - the immune cells and neutrophil all had "0" as a proportion. Did this line up with your result?

This is different than the expected result from the paper. Part of this could be the fact that the single cell data used in the tutorial was a subset, but it seems like I might be missing something else too..

[tjbutler003]

@solinvicta I also used what @tkamth proposed, but with my own data. In that, I had 3 groups, one of which was only one cell type, the other two which had 4 and 6 cell types in them. I made artificial bulk data using 1/11 * cell type (same proportion of every cell type for bulk). The group that had only one cell type had more variable results than the other cell types, which for the most part were very on-target with estimated proportion (1/11). I am wondering if this then has something to do with the way groups of one cell type are handled? @xuranw any suggestions?

[kabirium]

@solinvicta Could it be because as mentioned there is limited single-cell data uploaded on github that I too am having the same problem? To quote @xuranw "Due to the limited space of Github, we can only demo music_prop.cluster with a subset of mouse kidney single cell dataset. Therefore, the results might be different from the one presented in the paper due to incomplete reference single cell dataset." Also, I don't think @xuranw is active or interested in providing support in the forums.