plotOrdination, plotJointRPCA and modification to plotRDA

[TuomasBorman]

We have been using scater::plotReducedDim() to visualize reduced dimensionalities. However, it has couple limitations. For instance, it does not have functionality to add ellipses or to connect paired samples. The idea of plotOrdination() is to create a method that allows us to create easily visualizations commonly used in microbiome context. This can be easily extended later if needed.

I checked the literature, and identified the plots below. Is there other layouts that should be implemented?

data("Tengeler2020")
tse <- Tengeler2020
tse <- transformAssay(tse, method = "clr", pseudocount = TRUE)
tse <- scater::runPCA(tse, assay.type = "clr")

plotOrdination(tse, "PCA", colour.by = "patient_status")

plotOrdination(tse, "PCA", colour.by = "patient_status", fill.by = "patient_status", add.ellipse = TRUE)

plotOrdination(tse, "PCA", colour.by = "patient_status", add.points = FALSE, add.ellipse = TRUE)

plotOrdination(tse, "PCA", add.density = TRUE, colour.by = "patient_status")

plotOrdination(tse, "PCA", colour.by = "patient_status", add.centroids = TRUE)

plotOrdination(tse, "PCA", colour.by = "patient_status", add.centroids = TRUE, add.rotation = TRUE)

plotOrdination(tse, "PCA", colour.by = "patient_status", add.centroids.lines = TRUE)

plotOrdination(tse, "PCA", colour.by = "patient_status", add.centroids = TRUE, add.vectors = TRUE)

library(microbiomeDataSets)

mae <- microbiomeDataSets::peerj32()
tse <- getWithColData(mae, 1)
tse <- transformAssay(tse, method = "clr", pseudocount = TRUE)
tse <- scater::runPCA(tse, assay.type = "clr")

plotOrdination(tse, "PCA", colour.by = "group", pair.by = "subject")

plotOrdination(tse, "PCA", colour.by = "group", pair.by = "subject", sort.by = "time", shape.by = "sex")

[TuomasBorman]

@antagomir Any further ideas visualizing PCA/PCoA results?

[antagomir]

Seems great.

1. Explained percentages on the axes are a common need.
2. This should support, at least implicitly, other transformations besides CLR (e.g. robust CLR..) or other distances than Aitchison (e.g. robust Aitchison..)
3. Sometimes scaling of features is necessary (zero mean, unit variance) because this makes the loadings more directly comparable. To consider whether it is done before the function or as part of it.
4. Should be synchronized with plotLoadings()?

[TuomasBorman]

2.-3. This is working just like scater::plotReducedDim(). This does not do any calculations.

[antagomir]

2.-3. This is working just like scater::plotReducedDim(). This does not do any calculations.

Oh yes.

[TuomasBorman]

Remember to update plotRDA. Make sure that all options that are in plotReducedDim are also available in this new function.

Check also that the naming of arguments is harmonized. Currently, some of the arguments are passed to plotReducedDim without touching them, leading to usage of to naming conventions. For instance, this does not work

plotRDA(tse, dimred = "RDA", colour.by = "disease", shape.by = "gender")

but this works

plotRDA(tse, dimred = "RDA", colour.by = "disease", shape_by = "gender")

[TuomasBorman]

Add option for adding sample names:

add.colnames = FALSE

It can take either TRUE, FALSE or vector value that has the displayed sample names.

[TuomasBorman]

in plotRDA (and plotOrdination) make sure that axis labels are from reducedDim() results. Currently, plotRDA overwrites somewhere the axis labels. In the example below, y axis should be MDS1 as there is only one constrained axis (dbRDA1). However, they are now dbRDA1 and dbRDA2.

library(miaViz)

data(GlobalPatterns)

tse <- GlobalPatterns
tse <- tse[, tse[["SampleType"]] %in% c("Soil", "Feces")]
tse <- addRDA(tse, assay.type = "counts", formula = x ~ SampleType)
plotRDA(tse, "RDA")
# The y axis should be MDS1. Take names directly from reducedDim
colnames(reducedDim(tse, "RDA"))

[TuomasBorman]

Add support for joint-RPCA: microbiome/mia#808

Check also plotLoadings and joint-RPCA

[artur-sannikov]

We can add ellipses this way:

colnames(p$data)

# r$> colnames(p$data)
# [1] "X"         "Y"         "colour_by" "order_by" 

p <- plotReducedDim(tse, "MDS_Bray", color_by = "group") +
  stat_ellipse(aes(color = colour_by))

The ellipses functionality can be implemented like in the plotOrdination function, but it's not strictly necessary.

[TuomasBorman]

Thanks! This is very true, but requires knowledge on ggplot class, thus it is little bit "hacky" way

[TuomasBorman]

I removed matrix method.

This method just converts matrix to TreeSE first and then applies TreeSE-method.

This is just causing maintenance overhead without adding much value (the whole idea of the package is to provide SE-based methods)

[TuomasBorman]

These comes from plotOrdination in the new version

[TuomasBorman]

Lots of the functionality comes now from shared plotOrdination function.

[TuomasBorman]

Good to go.

This PR is expecting that changes from microbiome/mia#841 are merged. So that must be done first.