AutoSpectral: plotting
- olivertburton
- 10 hours ago
- 4 min read
In this article, let’s look at a couple of the functions that AutoSpectral offers for plotting your spectra and data. AutoSpectral isn't intended to replace flow analysis software like FlowJo or FCS Express, but it's essential to have some options for visualizing the outputs.
Let’s load in some spectra, here from OMIP-102.
s8.fcs <- "~/AutoSpectral_data/S8_data/BP0323502_1.fcs"
chorus.spectra <- read.bd.spectra( s8.fcs )
chorus.spectra[ 1:6, 1:5 ]
#> UV1 (375)-A UV2 (390)-A UV3 (420)-A UV4 (440)-A UV5 (460)-A
#> CD40 BUV395-A 0.15491615 0.43476247 1.0000000 0.7179330 0.36265173
#> CD3 BUV496-A 0.02629492 0.08357235 0.2908688 0.2602035 0.24447993
#> CD56 BUV563-A 0.04424246 0.10662948 0.2355437 0.1685517 0.08813200
#> CD141 BUV615-A 0.02715625 0.06347369 0.1510007 0.1212349 0.06775691
#> CD303 BUV661-A 0.01522760 0.04873425 0.1771819 0.1613903 0.09411419
#> CD86 BUV737-A 0.03180002 0.10011904 0.3424454 0.3125159 0.18939716The basic tools are spectral.trace, which gives line graphs of the spectra for each fluorophore in the matrix, and spectral.heatmap, which gives the same information as a heatmap.
spectral.heatmap( chorus.spectra, title = "OMIP102", color.palette = "mako",
show.legend = FALSE )
Note that you can set save = FALSE to just have the plots appear in the Viewer, but the saved versions are scaled automatically to try to adjust for the amount of data present.
The other option is a set of traces for each fluorophore.
spectral.trace( chorus.spectra, title = "OMIP102", split.lasers = FALSE,
show.legend = FALSE )
You can add the legend back:
spectral.trace( chorus.spectra, title = "OMIP102", split.lasers = FALSE,
show.legend = FALSE )
It’s also better to split it up by laser when there are many fluorophores:
spectral.trace( chorus.spectra, title = "OMIP102", split.lasers = TRUE )
In this case, the laser data are out of order because OMIP-102 was produced on an early version of the BD FACSDiscoverS8. I'll fix this eventually, but it hasn't been an issue on any other data I've tried unless the files have been corrupted exported by FlowJo first.
We can create a cosine similarity matrix plot in one of two ways.
cosine.similarity.plot( chorus.spectra, title = "Cosine_OMIP102",
output.dir = getwd() )
This way generates a square matrix (duplicated info across the diagonal) showing the cosine similarity values between each pair. The mixing matrix condition number is listed below for reference.
We can also call create.heatmap which is a more general-purpose function that’s more useful for things like plotting the “hotspot matrix”. This gives you a bit more control, but there are aspects that still need some work (sorry).
For this, let’s look at a small subset of the matrix (the first five fluorophores). We’ll calculate the cosine similarity matrix by calling cosine.similarity (hint, do the same with other functions to see other aspects of your data).
matrix.subset <- chorus.spectra[ 1:5, ]
small.cosine.matrix <- cosine.similarity( matrix.subset )
create.heatmap( small.cosine.matrix, number.labels = TRUE,
color.palette = "turbo", plot.dir = getwd() )
Let’s change a couple of things.
create.heatmap( small.cosine.matrix,
number.labels = TRUE,
legend.label = "Cosine Similarity",
triangular = TRUE,
color.palette = "mako",
plot.dir = getwd() )
We have the option to pass min and max scale settings, which is useful for comparing multiple sets of data on the same scale (not really relevant in this case, though).
create.heatmap( small.cosine.matrix,
number.labels = FALSE,
legend.label = "Cosine Similarity",
title = "OMIP102_Cosine_Similarity",
triangular = TRUE,
fixed.scale = TRUE,
scale.min = 0, scale.max = 2,
color.palette = "plasma",
plot.dir = getwd() )
Finally, we can make x-y biplots of FCS data, which can be useful for quickly checking what’s going on. To be fair, this is where programs with graphical user interfaces like FlowJo, FCS Express, SpectroFlo, etc. really excel, so doing it here is in R not really the best option.
To start, we can load in an FCS file–let’s use the OMIP-102 file–and we need to call up the AutoSpectral parameter list because I haven’t gotten around to removing that dependency yet.
asp <- get.autospectral.param( cytometer = "s8" )
omip102.file <- "~/AutoSpectral_data/S8_data/BP0323502_1.fcs"
omip102.ff <- suppressWarnings( flowCore::read.FCS( omip102.file,
transformation = FALSE,
truncate_max_range = FALSE,
emptyValue = FALSE ) )
omip102.data <- flowCore::exprs( omip102.ff )Now we can call the plotting function:
create.biplot( omip102.data,
x.dim = "BUV395-A", y.dim = "BUV496-A",
asp,
title = "OMIP102_biplot1" )
Note that this is currently ungated. You can play around with the automated gating functions in AutoSpectral if you want (these work okay with OMIP-102), but you’ll be better off using a graphical interface.
We can change the plot limits, color palette, etc. I haven’t implemented a rainbow palette like FlowJo has yet.
create.biplot( omip102.data,
x.dim = "BUV395-A", y.dim = "BUV496-A",
x.min = -15000, x.max = 1e6,
color.palette = "inferno",
asp,
title = "OMIP102_biplot2" )
Change the width basis parameter to compress the data more or less around zero. As we're using a replica of the FlowJo biexponential transform in R, the width basis caps out at -1000. I've rigged it so that putting in values less than -1000 will increase compression by altering the positive log decades, the other parameter you can use in FlowJo.
create.biplot( omip102.data,
x.dim = "BUV395-A", y.dim = "BUV496-A",
x.min = -15000, x.max = 1e6,
y.width.basis = -250,
color.palette = "inferno",
asp,
title = "OMIP102_biplot3" )
The maximum number of points to be plotted is capped by default at 1e5 to speed up the plotting. Setting an arbitrarily big number will plot everything.
create.biplot( omip102.data,
x.dim = "BUV395-A", y.dim = "BUV496-A",
x.min = -15000, x.max = 1e6,
y.max = 2e6,
max.points = 1e8,
color.palette = "viridis",
asp,
title = "OMIP102_biplot4" )
Comments