It's possible to impute/enhance peak signals for scATAC-seq dataset by signac ?

[PeggySze]

I'd like to draw a heatmap showing peak signals in differernt groups of cells. However, due to the sparsity of scATAC-seq data, it doesn't look good. It's possible to impute/enhance peak signals by smoothing signal across nearby cells in scATAC-seq dataset just like the way MAGIC(Markov Affinity-based Graph Imputation of Cells ) does in scRNA-seq ?

[timoast]

You could try running MAGIC, I haven't used it but it should also work for scATAC-seq

What I've used in the past is a simple k-nearest neighbor smoothing of the peak matrix for visualization of individual peaks in datasets that are very sparse, there's an example here: https://github.com/timoast/nanobody/blob/master/code/bmmc_dual/trajectory.R#L80-L91

[jiangpuxuan]

Hi,
I am also interested in this issue. Did MAGIC perform well in scATAC-seq analysis ?

[sjasws]

Dear @PeggySze @timoast @jiangpuxuan ,
Have you implemented MAGIC in signac to smooth the activity matrix?

[timoast]

no, and no plan to

[jiangpuxuan]

Dear @PeggySze @timoast @jiangpuxuan , Have you implemented MAGIC in signac to smooth the activity matrix?

Yes. It smoothed. But I think it's meaningless to MAGIC the Genescore matrix.

[sjasws]

Dear @PeggySze @timoast @jiangpuxuan , Have you implemented MAGIC in signac to smooth the activity matrix?

Yes. It smoothed. But I think it's meaningless to MAGIC the Genescore matrix.

How did you achieve that? I have now obtained the harmony integrated graph through the signac process and now try to smooth the inferred gene activity from peak on top of it, just as ArchR did. Because I found that gene activity was much better embedded on the umap after ArchR imputed by MAGIC, rather than the original data, which was very helpful in determining cluster subtypes. Now that you've done it, can you share the code or process? Are you smoothing based on your own graphics?

[jiangpuxuan]

Dear @PeggySze @timoast @jiangpuxuan , Have you implemented MAGIC in signac to smooth the activity matrix?

Yes. It smoothed. But I think it's meaningless to MAGIC the Genescore matrix.

How did you achieve that? I have now obtained the harmony integrated graph through the signac process and now try to smooth the inferred gene activity from peak on top of it, just as ArchR did. Because I found that gene activity was much better embedded on the umap after ArchR imputed by MAGIC, rather than the original data, which was very helpful in determining cluster subtypes. Now that you've done it, can you share the code or process? Are you smoothing based on your own graphics?

I use ArchR, extracting the GeneScoreMatrix and using it to make a seurat object.

mtx<-getMatrixFromProject(your_proj,useMatrix = "GeneScoreMatrix",)
 mtx@assays@data@listData$GeneScoreMatrix ->mtx.data
  mtx.data@Dimnames[[1]] <-mtx@elementMetadata$name
seuset.atac <- CreateSeuratObject(mtx.data)

new.ident<-your_proj$yourClusters
new.ident %>% head()
your_proj$cellNames %>% head()

names(new.ident)<-your_proj$cellNames
new.ident<-as.factor(new.ident)
new.ident %>% unique()

[email protected]<-new.ident

seuset.atac<-magic(seuset.atac)

[sjasws]

Dear @PeggySze @timoast @jiangpuxuan , Have you implemented MAGIC in signac to smooth the activity matrix?

Yes. It smoothed. But I think it's meaningless to MAGIC the Genescore matrix.

How did you achieve that? I have now obtained the harmony integrated graph through the signac process and now try to smooth the inferred gene activity from peak on top of it, just as ArchR did. Because I found that gene activity was much better embedded on the umap after ArchR imputed by MAGIC, rather than the original data, which was very helpful in determining cluster subtypes. Now that you've done it, can you share the code or process? Are you smoothing based on your own graphics?

I use ArchR, extracting the GeneScoreMatrix and using it to make a seurat object.

mtx<-getMatrixFromProject(your_proj,useMatrix = "GeneScoreMatrix",)
 mtx@assays@data@listData$GeneScoreMatrix ->mtx.data
  mtx.data@Dimnames[[1]] <-mtx@elementMetadata$name
seuset.atac <- CreateSeuratObject(mtx.data)

new.ident<-your_proj$yourClusters
new.ident %>% head()
your_proj$cellNames %>% head()

names(new.ident)<-your_proj$cellNames
new.ident<-as.factor(new.ident)
new.ident %>% unique()

[email protected]<-new.ident

seuset.atac<-magic(seuset.atac)

ArchR has its own impute feature, so you can implement it directly on ArchR. I am very inspired by you, if pass data from signac to ArchR, it seems that it can try to implement features that ArchR has but signac does not...