Genes returned by getMarkers() vs markerHeatmap()

[oligomyeggo]

Hello, and thank you for an incredible package to analyze scATAC-seq data! I have been following your detailed manual with my own data and it has been working great. I have a question about some of the data generated from section 7.3,, Identifying Marker Genes, namely the output from running getMarkers() and markerHeatmap(). Both functions take in the SummarizedExperiment object generated from running the getMarkerFeatures() function, and for both functions I am using the same cutOff values.

From my understanding, the DataFrame objects returned from running getMarkers() returns a DataFrame for each cluster with marker genes sorted by FDR. I am struggling to understand how the top n genes are selected per cluster by the markerHeatmap() function, though. They do not match the order of the genes in the data frames (sorted by FDR), and they also don't seem to match the order of the genes in the data frames if I sort by other metrics (such as MeanDiff or Log2FC). Also, for one of my clusters, I am only getting back one gene that matches the cutoff values after running getMarkers(). However, I am getting n=15 genes back for this cluster when running markerHeatmap().

I see in the code that both getMarkers() and markerHeatmap() start off by generating the same passMat object, but that the passMat is manipulated a little more in the markerHeatmap() function and is used to generated a spmat object which is ultimately used to get the heatmap marker genes. I think I am just missing something in the markerHeatmap() code, but any insight into how the top marker genes for the markerHeatmap() are generated and why they differ from the sorted data frames generated by getMarkers() would be appreciated. Thank you!

ArchR version = 1.0.2

[rcorces]

the spmat line that you are referring to is this:

ArchR/R/MarkerFeatures.R

Line 969 in f6c0388

spmat <- passMat / rowSums(passMat)

I believe that this is just a way to measure how "unique" a peak is to a given cell grouping. A row is a gene and by normalizing by the rowSums you get how many cell groups have this marker in passMat. The markers get ordered by this normalized value which doesnt really have any bearing on the log2FC etc.