Scatterplots to compare two data sets in R

Question

I would like to compare two datasets of the same length using scatterplots and then colour the differentially expressed genes based on the list of genes with specified p-values. To be more specific it looks something like below :

enter image description here

Replacing the x-axis with data A and y-axis with data B and Significant P value <0.05) probes are shown in blue.

 A = rnorm(100) ### datset A

 B = rnorm(100) ### datset B 

 p = rnorm(100) ### p-values

I tried to reproduce the same kind of plot by first producing a normal scatterplot . The data I originally have is for 15000 probes.

       col = rep('black', length(A))
       col[p< 0.05] <- 'blue'
       plot(A, B, col=col)

enter image description here

What have you tried? Please give us an [reproducible example](http://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example). — sgibb, Sep 03 '13 at 11:11
@ sgibb: If you mean dput of the data, it was quite large and did not want to crowd it up here. The code I tried using was similar to the above but I only want to have an idea as to how its done. — user1805343, Sep 03 '13 at 11:51
The black/blue coloring by p-value looks good. What are you stuck on? Also, when generating random numbers, try using `set.seed(1)` (or any other number) so the results are fully reproducible. — BenBarnes, Sep 03 '13 at 12:36
@ Ben Barnes : Sure . I was able to complete it except am unable to produce the red lines shown in the plot. — user1805343, Sep 03 '13 at 13:12
I suspect the red lines correspond to whatever statistic the p-values are based on. Could you clarify how p-values were derived? — BenBarnes, Sep 03 '13 at 13:41

score 3 · Accepted Answer · answered Sep 03 '13 at 13:13

Deducing from the fact that the axis ranges are [0,1] in the original figure, I take it you want to plot a scatterplot of p-values under two conditions and two sets of tests.

Here's how the plotting itself is done:

# Generate an artificial dataset
library(MASS)
set.seed(1)

# Suitably chosen covariance matrix
covariancematrix <- matrix(c(0.02,0.019,0.019,0.02), nrow=2)
#> print(cov2cor(covariancematrix))
#     [,1] [,2]
#[1,] 1.00 0.95
#[2,] 0.95 1.00

# Randomize 20,000 observations and constraint them to a p-value like range
pvalues <- mvrnorm(20000, mu=c(0.5,0.5), Sigma=covariancematrix)
colnames(pvalues) <- c("p value condition 1", "p value condition 2")
rownames(pvalues) <- paste("Probe/gene id", 1:20000)
# p-values should be within range [0,1]
pvalues[pvalues<0] <- 0
pvalues[pvalues>1] <- 1

#> str(pvalues)
# num [1:20000, 1:2] 0.461 0.54 0.52 0.518 0.61 ...
# - attr(*, "dimnames")=List of 2
#  ..$ : chr [1:20000] "Probe/gene id 1" "Probe/gene id 2" "Probe/gene id 3" "Probe/gene id 4" ...
#  ..$ : chr [1:2] "p value condition 1" "p value condition 2"
#> head(pvalues)
#                p value condition 1 p value condition 2
#Probe/gene id 1           0.4614707           0.4767356
#Probe/gene id 2           0.5398754           0.5583752
#Probe/gene id 3           0.5196980           0.5251556
#Probe/gene id 4           0.5182167           0.4471374
#Probe/gene id 5           0.6097540           0.5745387
#Probe/gene id 6           0.4652409           0.3940416

# The plotting call itself
plot(   # If this is a 2-column matrix, then first column pvalues[,1] will be by default the x-axis and second column pvalue[,2] the y-axis
    # Can be a matrix with 2 columns or a data.frame with 2 columns
    x = pvalues, 
    # Analogous if you have two separated vectors instead of two columns, change to something like:
    # x = pvalues[,1], # x-axis observation values
    # y = pvalues[,2], # y-axis observation values
    # x- and y-axis ranges [0,1]
    xlim=c(0,1), 
    ylim=c(0,1), 
    # Select filled dots as the symbols
    pch=16, 
    # Conditional color vector based on where the observation is located
    col=apply(pvalues, MARGIN=1, FUN=function(x) { ifelse(1*x[1] +0.05 > x[2] & 1*x[1] -0.05 < x[2], 
        # Color for dots inside the area (semi-transparent black)
        "#00000075", 
        # Color for dots outside the area (semi-transparent blue)
        "#0000FF75") 
        } ),
    # Axis labels
    xlab="p values in differences condition 1", 
    ylab="p values in differences condition 2"
    )
# Draw the lines, formula:
# y = 1*x + 0.05
abline(a=0.05, b=1, col="red", lwd=2)
# y = 1*x - 0.05
abline(a=-0.05, b=1, col="red", lwd=2)

Here's how it turns out:

enter image description here

If you are yet to have the p-values for the differences, then you need to perform suitable statistical testing first, for example RankProduct.

Scatterplots to compare two data sets in R

1 Answers1