Inconsistent results from pseudo-time heatmap and UMAP embedding of peaks

[callmeracle]

Hi,
Recently, I used the ArchR package to process my scATAC-seq data and my goal was to investigate the openness dynamics of 27 custom-defined hotspots (not identified by MACS2) during differentiation. To achieve this, I generated pseudotime heatmaps (see attached file 1) and visualized hotspot openness using UMAP embeddings (attached file 2, only show one of the hotspots as an example). However, I observed an inconsistency between the openness changes displayed in the pseudotime heatmaps and those reflected in the UMAP embeddings. Notably, the UMAP embeddings were derived from the TileMatrix rather than the PeakMatrix.

I would greatly appreciate your insights on how to resolve this discrepancy. Attached, you will find the heatmap and UMAP embedding plot for reference.

proj_SC <- addIterativeLSI(
    ArchRProj = proj_SC,
    useMatrix = "TileMatrix",
    name = "IterativeLSI_SC",
)
proj_SC <- addHarmony(
    ArchRProj = proj_SC,
    reducedDims = "IterativeLSI_SC",
    name = "Harmony",
    groupBy = "group",
force= TRUE
)
proj_SC <- addClusters(
    input = proj_SC,
    reducedDims = "Harmony",
    method = "Seurat",
    name = "Clusters_SC",
    resolution = 0.8,
force=TRUE
)
proj_SC <- addUMAP(
    ArchRProj = proj_SC,
    reducedDims = "Harmony",
    name = "UMAP_SC",
	force =TRUE
)

hotspot = read.table("Hotspot.bed")
hotspot.gr = GRanges(hotspot[,1],IRanges(hotspot[,2], hotspot[,3]), name=hotspot[,4])
proj_SC <- addPeakSet(proj_SC, peakSet = hotspot.gr, force=TRUE)
proj_SC <- addPeakMatrix(proj_SC)
proj_SC <- addImputeWeights(proj_SC)

trajectory <- c("Self renewing","Quiescent","Activated/Proliferating","Differentiate 1","Differentiate 2")
proj_SC <- addTrajectory(
    ArchRProj = proj_SC,
    name = "traj",
    groupBy = "Clusters_SC",
    trajectory = trajectory,
    embedding = "UMAP_SC",
    force = TRUE
)

traj_hotspot  <- getTrajectory(ArchRProj = proj_SC, name = "traj", useMatrix = "PeakMatrix", log2Norm = TRUE,trajectoryLabel = "group")
p4 <- plotTrajectoryHeatmap(traj_hotspot,varCutOff=0, pal = paletteContinuous(set = "solarExtra"),colorColumns=TRUE,columnPal = paletteDiscrete(values = unique(colData(traj_hotspot)$label)))

p <- plotTrajectory(
      proj_SC,
      trajectory = "traj",  
      colorBy = "PeakMatrix",
      name = "P2",   
      continuousSet = "horizonExtra",
      embedding = "UMAP_SC",    
      plotAs = "points",
      size = 2,
      addArrow = FALSE
    )

Trajectory_heatmap_hotspot_clusters.pdf

P2.pdf

[immanuelazn]

Hi @callmeracle! Thanks for using ArchR! Please make sure that your post belongs in the Issues section. Only bugs and error reports belong in the Issues section. Usage questions and feature requests should be posted in the Discussions section, not in Issues.

If you are getting an error, it is likely due to something specific to your dataset, usage, or computational environment, all of which are extremely challenging to troubleshoot. As such, we require reproducible examples (preferably using the tutorial dataset) from users who want assistance. If you cannot reproduce your error, we will not be able to help.
Before going through the work of making a reproducible example, search the previous Issues, Discussions, function definitions, or the ArchR manual and you will likely find the answers you are looking for.
If your post does not contain a reproducible example, it is unlikely to receive a response.

In addition to a reproducible example, you must do the following things before we help you, unless your original post already contained this information:
1. If you've encountered an error, have you already searched previous Issues to make sure that this hasn't already been solved?
2. Did you post your log file? If not, add it now.
3. Remove any screenshots that contain text and instead copy and paste the text using markdown's codeblock syntax (three consecutive backticks). You can do this by editing your original post.

[Pirhhot]

Hello,

I am not in the archr team but my guess would be the scaling of the rows performed in the heatmap.
Each row is scaled then "hard clipped" for better visualization and clustering of the rows.
Maybe try with scaleRows = FALSE to check or with softer limits params of (2,2) (what is most often used in papers as it more or less corresponds to removing 5% of the cells).
Also, have you tried visualizing your pseudotime values on the umap ?

Good luck

[callmeracle]

Dear Pirhhot,

Thank you for your helpful suggestions, and I apologize for my delayed response. After implementing your recommendations by setting scaleRows = FALSE and adjusting the softer limits parameters, I observed a significantly improved association in the trajectory heatmap visualization. Additionally, I mapped the pseudotime values onto the UMAP projection, and the resulting pattern of openness changes along the trajectory aligns well with the trends observed in my initial analysis.

Best,
Qiang