What is the meaning density in CummeRbund plots?

[Carlos Borroto]

Hi,

I recently started using CummeRbund for downstream analysis of cuffdiff output. Really nice tool BTW. I can't figured out exactly how to read the density plot:
http://compbio.mit.edu/cummeRbund/manual.html#Global

Any help on this? What is the meaning of density in this context?
Thanks,
Carlos

[lgoff]

Density plot

Hi Carlos,
You can think of the density as a 'smoothed' representation of a histogram. In this particular case, the csDensity plot is being used to examine the distribution of FPKM values across individual samples.

Check out http://en.wikipedia.org/wiki/Kernel_density_estimation for more specific information on what is being represented.

Thanks for trying cummeRbund!

Cheers,
Loyal

[Carlos Borroto]

Thanks for the answer, I understand now.

Looking forward for the cookbook recipes.

[apadr007]

Since we're on the subject, what is the criteria for something having significance in volcano plots?

I noticed that in some of my graphs my p-value may be less than 0.05 and not be significant according to whats plotted. Can you please elaborate on this.

Much appreciated,

[lgoff]

Since we're on the subject, what is the criteria for something having significance in volcano plots?

I noticed that in some of my graphs my p-value may be less than 0.05 and not be significant according to whats plotted. Can you please elaborate on this.

Much appreciated,

Hi apadr007,
The significance calls in cummeRbund were originally done against the 'significant' output field from cuffdiff. This was across the board as we were letting cuffdiff decide on a significance threshold. We have since implemented the MTC in cummeRbund and you can specify your own alpha for calls to 'getSig()' for example. However, to your point, I have neglected to include this in the csVolcano plotting methods. So as of now, it's still plotting on the original 'significant' field from cuffdiff. I will make this change for a future version such that you can either turn off the significance callouts and/or provide an alpha value to csVolcano on which to draw your significance threshold.

Thanks for pointing this out!

Cheers,
Loyal

[billstevens]

Hi Carlos,
You can think of the density as a 'smoothed' representation of a histogram. In this particular case, the csDensity plot is being used to examine the distribution of FPKM values across individual samples.

Check out http://en.wikipedia.org/wiki/Kernel_density_estimation for more specific information on what is being represented.

Thanks for trying cummeRbund!

Cheers,
Loyal

Hi,
I also have a quick question on the graph. Since its a log graph, does this graph just not show FPKM values less than 1? One of the reasons I am wondering is because my probability density function (just for from eyeing it) looks like it is less than 1.

[apadr007]

Thank you, Loyal. Also, how can one generate a table based on the data that csVolcano is plotting? The reason I ask is because after manually inspecting my cuff diff output I am not seeing some of the extremely high points csVolcano is plotting. So is there a way to produce a txt file from csVolcano as well as its graph?

[sdriscoll]

Hi,
I also have a quick question on the graph. Since its a log graph, does this graph just not show FPKM values less than 1? One of the reasons I am wondering is because my probability density function (just for from eyeing it) looks like it is less than 1.

I think in the log plots where they are plotting FPKM values they add 1 to the data. If you don't the density plot looks totally different.

try this in R (assuming you imported your data to a variable 'cuff' and your samples are named 'sample1' and 'sample2').

plot(density(diffData(cuff@genes,"sample1","sample2")$value_1))

verses

plot(density(diffData(cuff@genes,"sample1","sample2")$value_1+1))

R's kernel density algorithm extends the smoothed line past the actual extent of the data its smoothing. So you'll see the density line go below 1 even if 1 is the lowest value.

[lgoff]

Hi,
I also have a quick question on the graph. Since its a log graph, does this graph just not show FPKM values less than 1? One of the reasons I am wondering is because my probability density function (just for from eyeing it) looks like it is less than 1.

Hi Bill,
For the csDensity plot, (and in general), when logMode=T then a pseudocount (accessible as 'pseudocount' argument and default=1) is added prior to log transformation.

Loyal

[lgoff]

Thank you, Loyal. Also, how can one generate a table based on the data that csVolcano is plotting? The reason I ask is because after manually inspecting my cuff diff output I am not seeing some of the extremely high points csVolcano is plotting. So is there a way to produce a txt file from csVolcano as well as its graph?

The csVolcano is generated from data that can be retrieved by using the diffData() method around a CuffData, CuffGeneSet, or CuffGene object:

e.g.

myDiff<-diffData(genes(cuff))

Cheers,
Loyal

[billstevens]

How do you change the plotting parameters (title, x-axis, sample names, etc.) in csDensity? I tried to put it in the call, but that didn't work:

densityplot1 <- csDensity(genes(cuffy), xlab="xaxis")

When I then ran densityplot1, it didn't change the name of the x-axis.

I also tried just adjusting it after it plotted, but I kept getting an error saying plot.new has not been created.

[sdriscoll]

They are using the ggplot2 package. I came across a short tutorial for it somewhere. You can't use they typical R options but you can edit the plots after they are generated. I don't remember the. Commands though. Google that package and look for their tutorial.

[lgoff]

I think in the log plots where they are plotting FPKM values they add 1 to the data. If you don't the density plot looks totally different.

Hi All, this is correct. In general when we have to log-transform, we add a pseudocount and we default to using '1'. When we do this however, we always add a 'pseudocount' argument that you can change to whatever value you would prefer to add.

Cheers
Loyal

[ftorri]

Hi,

I am trying to pull out from the scatter plot the list of the outliers, but I haven't found a way through. Does anyone of you have any suggestions?

Thanks!
Fed

[apadr007]

^ This would be helpful ^