Gene Integration Matrix of Genes from scRNA

[abbaslab]

I used addGeneIntegrationMatrix as per the manual online. I want to output the table of the GeneIntegrationMatrix with the rows being genes and cell names as columns.

Can you please confirm if this is the correct approach?

x<-getMatrixFromProject(tumor_all_Clusters, "GeneIntegrationMatrix")
expression <- x@assays@data$GeneIntegrationMatrix # this generated a matrix of expression levels but without row names 
rownames(expression) <- x@elementMetadata$name #here, I am assuming that the order of the row names is maintained

1. Are these the correct steps?
2. Are the expression level in the "expression" matrix representing an inferred level of scRNA ? Or is that the exact value from the corresponding matched scRNA from the seurat object?
3. When running length(unique(x@colData$predictedCell)), we end up having 2472 cells being used from the scRNA object (original is >60,000 cells) to match the 39604 cells from my scATAC object. Is that expected that only a subset of cells from the scRNA object end up accounting for the predicted expression?

In a similar line of question regrind the addGeneIntegrationMatrix step:
for the groupRNA option which uses a prior knowledge of the scRNA format, I am not sure if my approach was correct:

1. My scATAC ArchR object has 39604 tumor cells from 8 patients (22 timepoints). In the analysis of the scATAC, I am doing the analysis "By Cluster" and not by patient.
2. In the groupRNA in the addGeneIntegrationMatrix I defined the patient groups (i.e. the 8 patients).
   In the example on ArchR, cell types were used whereby things could be distinguished by the biology. but for purpose of tumor cells, I cant define the corresponding cluster in the scRNA therefore used the patient level for the integration. Is this expected to affect the results of the integration? Technically, most of my scATAC clusters seem to be clustering by patient despite batch correction.

Thank you

[rcorces]

Are these the correct steps?

Seems fine to me

Are the expression level in the "expression" matrix representing an inferred level of scRNA ? Or is that the exact value from the corresponding matched scRNA from the seurat object?

This is the value from the scRNA data.

When running length(unique(x@colData$predictedCell)), we end up having 2472 cells being used from the scRNA object (original is >60,000 cells) to match the 39604 cells from my scATAC object. Is that expected that only a subset of cells from the scRNA object end up accounting for the predicted expression?

This can happen. It is obviously not ideal but it is one of the pitfalls of integration. The developers of Seurat, which is used for the integration, would be better positioned to comment further.

My scATAC ArchR object has 39604 tumor cells from 8 patients (22 timepoints). In the analysis of the scATAC, I am doing the analysis "By Cluster" and not by patient.
In the groupRNA in the addGeneIntegrationMatrix I defined the patient groups (i.e. the 8 patients).

I am not sure I understand your question. You can constrain the integration however makes the most sense for your data. If you have acquired snATAC and sc/snRNA from the same samples, then you should probably constrain by sample, not by cluster.

[abbaslab]

Thanks Ryan- I think you answered my questions, apologies for lack of clarity on last point but I think your response clarifies it.

One other related question: for the RNA expression matrix that is being extracted for the subset of 'integrated' cells from scRNA/scATAC: what would you recommend as a good test to assess differential expression in that case? Can I feed that data directly and do fold changes between groups/clusters? or does it have to go through levels of normalization again etc to do any expression analysis.

[rcorces]

I have some hesitations about doing differential expression calculations from linked data. I would defer to the authors of Seurat.

[janekim319]

What part of the full manual can I find information on getMatrixFromProject()? Thanks!

[rcorces]

Not every function is documented in the manual. Use the function-specific documentation:
https://www.archrproject.com/reference/index.html