Splitting columns in plotExprHeatmap by condition/metavariable

[JonasDBM]

Hi Helena,

thank you for this fantastic package and continuous support for the community!

Is there a way to use or modify the plotExprHeatmap function and for example plot two columns column of a specific marker split by condition/metavariable so that they are plotted next to each other?

Many thanks!

[HelenaLC]

Hey! Thanks for the kind words. -- Sorry, I'm not sure I understand exactly what you mean; could you be more specific? I assume section 4.4, 3rd example in the vignette is not what you mean

[JonasDBM]

Hi!
Sorry, my explanation was not very good!
Basically a pseudobulk expression heatmap with the cluster annotation on the y axis and the markers on the x axis where you have x columns per condition for a specific marker similar to this: [https://divingintogeneticsandgenomics.com/post/how-to-make-a-multi-group-dotplot-for-single-cell-rnaseq-data/]

[JonasDBM]

Sorry again, without the average expression component. Don`t think it makes a lot of technical sense for flow cytometry...

[HelenaLC]

You mean that blue/red dot plot right?

[JonasDBM]

Yes, like the 2nd heatmap of the vignette 4.4 but where each column of type_marker is duplicated and one column shows the pseudobulk expression of control the other stimulated condition.

[HelenaLC]

Short answer: No, can't do this in one line with CATALYST... But: the package builds around the SingleCellExperiment class, for which there is a ton of infrastructure... It'd be generally a good idea to become a little familiar with other package's capabilities. E.g., combining scuttle and scater one can do a lot of custom plots - I reproduced the one you sent in about 5', see code below. I can imagine you could write a lot of little wrappers for specific visualizations you'd like to see but are missing, and reuse them in various places. Sorry that it's perhaps a bit more complicated than expected :( but hope it helps :)

library(CATALYST)
library(scuttle)
library(scater)

data(PBMC_fs, PBMC_panel, PBMC_md)
sce <- prepData(PBMC_fs, PBMC_panel, PBMC_md)
sce <- cluster(sce, verbose=FALSE)

# compute cluster-condition expression medians using 'scuttle'
ids <- DataFrame(gid=sce$condition, kid=cluster_ids(sce, "meta7"))
pbs <- aggregateAcrossCells(sce, ids=ids, use.assay.type="exprs", statistics="median")

# add mock variable (cluster+group)
pbs$kid.gid <- with(colData(pbs), paste(kid, gid, sep="_"))

# plot using 'scater' (could also use 'plotHeatmap' here instead)
plotDots(pbs, 
    group="kid.gid",
    exprs_values="exprs",
    features=rownames(pbs),
    center=TRUE, scale=TRUE) +
    # aesthetics (could be improved further)
    coord_equal() +
    scale_size_continuous(range=c(0, 4)) +
    scale_color_gradient2(low="blue", high="red") +
    theme_minimal(6) + theme(axis.text.x=element_text(angle=45, hjust=1))

[HelenaLC]

Just to add one more example; once one gets the hang of this, the possibilities are endless... E.g., here I aggregated by cluster-sample, facet by cluster, showing the expression across all samples (rather than aggregating across them)

# compute cluster-sample expression medians using 'scuttle'
ids <- DataFrame(sid=sce$sample_id, kid=cluster_ids(sce, "meta7"))
pbs <- aggregateAcrossCells(sce, ids=ids, use.assay.type="exprs", statistics="median")

# add mock variable (cluster+sample)
pbs$kid.sid <- with(colData(pbs), paste(kid, sid, sep="_"))

# plot using 'scater' (could also use 'plotHeatmap' here instead)
p <- plotDots(pbs, 
    group="kid.sid",
    exprs_values="exprs",
    features=rownames(pbs),
    center=TRUE, scale=TRUE) +
    # aesthetics (could be improved further)
    scale_size_continuous(range=c(0, 4)) +
    scale_color_gradient2(low="blue", high="red") +
    theme_bw(6) + theme(axis.text.x=element_text(angle=45, hjust=1))

# add all metadata
i <- match(p$data$Group, pbs$kid.sid)
j <- setdiff(names(colData(pbs)), names(p$data))
p$data <- cbind(p$data, colData(pbs)[i, j])

p + facet_wrap(~kid, nrow=1, scales="free_x")

(could be made prettier of course...)