representing PCs as marginal distribution on x and Y acis

I came across a figure attached herewith and want to try the plotting style. But I can't get the grasp of what units are plotted as marginal distribution on X and Y. I have PCA plot and PC loadings, but how to plot them as this marginal distribution? Can someone explain? For the clarity colour codings in blue, cyan green and yellow are different samples and control, CS are treatment groups

enter image description here

pcatool deseq2 pcaplot pca

• 1,404 views • link

updated 2.4 years ago by Papyrus

• written 2.4 years ago by Yogesh •

1 answer

There are no "additional" units. For a PCA plot you do a scatter plot of the PC values (e.g. PC1 and PC2) on the X and Y axes, and the marginal plots are just the densities, histograms of those X and Y values that you plot in the scatter plot. You can try the ggMarginal function from the ggExtra package (see this page for examples) to produce a scatter plot with marginal distribution plots.

• 0 views • link

written 2.4 years ago by Papyrus

hi thank you for the response. i tried it but the histograms do not have metadata information like group type and sample type. how do i add that

below is the code that i am using..

plotPCA <- function (object, intgroup=c("condition"), ntop = 500, returnData = FALSE) 
{
   .
.
.
.

    p <- ggplot(data = d, aes_string(x = "PC3", y = "PC4", color = "group")) + 
        geom_point(size = 3) + xlab(paste0("PC3: ", round(percentVar[3] * 
        100), "% variance")) + ylab(paste0("PC4: ", round(percentVar[4] * 
        100), "% variance")) + coord_fixed()
       p_with_marginals <- ggMarginal(p, type = "histogram",groupfill=TRUE)

  print(p_with_marginals)

}

print(plotPCA(vsd, intgroup=c("Group", "type")))

• 0 views • link

written 2.4 years ago by Yogesh •

Here's some reproducible code to show the input data:

library(ggplot2)
library(ggExtra)

df <- matrix(rnorm(600),nrow = 50, ncol = 12)
group1 <- rep(c("a","b"), each = 6)
group2 <- rep(c("one","two","three"),each = 4)

pca <- prcomp(t(df), center = T, scale. = T)
varexp <- (pca$sdev^2 / sum(pca$sdev^2)) * 100

data <- data.frame(PC1 = pca$x[,1],
                   PC2 = pca$x[,2],
                   group1 = group1,
                   group2 = group2)

Then, an initial ggMarginal plot would me something like this:

plot1 <- ggplot(data, aes(y = PC2, x = PC1, color = group1, shape = group2)) +
  geom_point() + xlab(paste0("Dim1 (var: ",round(varexp[1],1),"%)")) +
  ylab(paste0("Dim2 (var: ",round(varexp[2],1),"%)")) + theme_bw()
plot.m1 <- ggMarginal(plot1, type = "density", groupFill = T, groupColour = T)
plot.m1

If you want to have different marginal distributions on both axes, you can do a solution such as proposed here (there may be other solutions for this):

plot1 <- ggplot(data, aes(y = PC2, x = PC1, color = group1, shape = group2)) +
  geom_point() + xlab(paste0("Dim1 (var: ",round(varexp[1],1),"%)")) +
  ylab(paste0("Dim2 (var: ",round(varexp[2],1),"%)")) + theme_bw()
plot.m1 <- ggMarginal(plot1, type = "density", groupFill = T, groupColour = T)

plot2 <- ggplot(data, aes(y = PC2, x = PC1, color = group2, shape = group2)) +
  geom_point() + xlab(paste0("Dim1 (var: ",round(varexp[1],1),"%)")) +
  ylab(paste0("Dim2 (var: ",round(varexp[2],1),"%)")) + theme_bw()
plot.m2 <- ggMarginal(plot2, type = "density", groupFill = T, groupColour = T)


plot.m1$grobs[plot.m1$layout$name == "topMargPlot"] <- 
  plot.m2$grobs[plot.m2$layout$name == "topMargPlot"]

plot.m1

• 0 views • link

written 2.4 years ago by Papyrus