Question

Jitter p-values in volcano plot

0

Entering edit mode

philipp.rewe ▴ 20

@philipprewe24159

Last seen 3.5 years ago

Würzburg

Hi, I working with a DESeq2 data set and I am using the package EnhancedVolcano to visualize my data. This is one example:

enter image description here

As you can see some of the genes have such a low p-value that DESeq2 places 0. This does not look very nice in the plot - said my PI. I guess there are two ways to address this issue (which is not really an issue rather than an aesthetic thing): Either in DESeq2 by calculating the exact p-value if possible or by placing randomly very low p-values or in EnhancedVolcano by applying some kind of jitter function to the values of the y-axis. I know that EnhancedVolcano is based on ggplot2 and there is the function called geom_jitter() but I am not able to apply it to my data.

Please let me know if you need more information or data. Thank you guys in advance!

Cheers, Philipp

EnhancedVolcano DESeq2 • 3.6k views

ADD COMMENT • link updated 3.5 years ago by Michael Love 41k • written 3.5 years ago by philipp.rewe ▴ 20

score 2 · Accepted Answer · 2020-11-06

2

Entering edit mode

Michael Love 41k

@mikelove

Last seen 9 hours ago

United States

I've used jitter for very small p-values before, see Figure 4 here:

Because DEXSeq outputs an adjusted p-value of 0 for some of the genes, we set these instead to a jittered value around 10−20, so that their number and location on the x-axis could be visualized. These jittered values should only be used for visualization.

https://f1000research.com/articles/7-952#f4

ADD COMMENT • link 3.5 years ago Michael Love 41k

1

Entering edit mode

Hey, just a quick feedback. I managed by following the description in the link: enter image description here

# subsetting rows with p-value 0:
pval_0 <- subset(res, res$pvalue == 0)
# creating a random list of numbers around 1e-300. 
pval_jitter <- round(rnorm(nrow(pval_0), mean =295, sd = 5),0) 
pval_jitter <- 1*10^-pval_jitter
# replace old p-values by new values
pval_0$pvalue <- pval_jitter
# creating a new data set for visualisation 
resA_1 <- subset(res,res$pvalue!=0)
resA_2 <- rbind(resA_1, padj0)

I guess it can be done in a more elegant way. But it does the job for me. I also noticed that numbers lower than 1e-320 or so are automatically set to 0. I do not know why.

Again, thank you so much for your help and these great packages!

Cheers, Philipp

ADD REPLY • link 3.5 years ago philipp.rewe ▴ 20

1

Entering edit mode

At some point R will represent a small p-value with 0.

!> t.test(rnorm(1000), rnorm(1000,3))$p.value
 [1] 0

ADD REPLY • link 3.5 years ago Michael Love 41k

0

Entering edit mode

Just for the EnhancedVolcano part, this may very well work and is worth a try:

EnhancedVolcano(...) + ggplot2::geom_jitter()

ADD REPLY • link 3.5 years ago Kevin Blighe ★ 3.9k

0

Entering edit mode

Thanks for your quick response. I tried that already but I did not really understand what actually happened in the resulting plot: enter image description here

ADD REPLY • link 3.5 years ago philipp.rewe ▴ 20

1

Entering edit mode

Ah, I see what happened there - geom_jitter() just adds a default black dot over the original geom points. You may need something like:

EnhancedVolcano(...) + ggplot2::geom_point(colour = 'white') + ggplot2::geom_jitter()

We are somewhat getting into a rabbit hole, though. If your supervisor is merely worried about aesthetics, then I would respond with this: good science and data does not have to be aesthetically pleasing.

ADD REPLY • link 3.5 years ago Kevin Blighe ★ 3.9k

1

Entering edit mode

haha, I will relay it to him.

ADD REPLY • link 3.5 years ago philipp.rewe ▴ 20

0

Entering edit mode

Great, thanks a lot! I might not be able to apply that to my data though but I will give it a try.

ADD REPLY • link 3.5 years ago philipp.rewe ▴ 20