correlation value in scatterplots using cummerbund??

[Marcos Lancia]

Hi,
I'm running cummerbund and I made a scatterplot of my data, but I'd like to see the correlation value of that plot. Is it possible? Is there any command to see the R value of that correlation?? Thanks!

[blakeoft]

You might have to use cor(), a base R function. However, I don't know exactly how to subset a CuffData object so that you get the proper input for cor().

[Marcos Lancia]

That's my big problem right now. I'd like to know how can I use cor() with CuffData.

[cmbetts]

That's my big problem right now. I'd like to know how can I use cor() with CuffData.

It's been a while since I've used cummeRbund, but I remember that you can use the fpkmMatrix() function (can't remember the exact usage) to get a matrix that plays nicely with the base R functions.

[blakeoft]

Thanks for the tip cmbetts. It looks like you want to execute

m <- fpkmMatrix(genes(cuffdiff_output))
cor(m[, 1], m[, 2]) # or whatever columns you need

Keep in mind that csScatter() might discard some of the data points. You can verify this by running

csScatter(genes(cuffdiff_output))
# compare to
plot(m[, 1], m[, 2])

[Marcos Lancia]

I'm progressing, now a new problem

Thanks for writing everyone! I finally could see the correlation value, but I crashed with another new problem. One of the r value is near to 0, but the plot looks very good, near to 1. I'm pretty sure that data plotted is the same analized. Anybody saw something similar? What did you do? Thanks!

[blakeoft]

Marcos,

Did you ever sort this out? I would double check that you're supplying the right vectors to cor(). It might also have to do with how csScatter doesn't plot all of the data points. It's hard for me to think of anything else since I don't have the data to play with myself.

[Marcos Lancia]

Hi, again

For example: I want to analyze TRAP_Sm_rep1 vs TRAP_Sm_rep2. So, I write:

>samples (cuff)

sample_index sample_name sample_name parameter value
1 1 SN16K_mock_rep2 <NA> <NA> <NA>
2 2 SN16K_mock_rep1 <NA> <NA> <NA>
3 3 SN16K_Sm_rep2 <NA> <NA> <NA>
4 4 SN16K_Sm_rep1 <NA> <NA> <NA>
5 5 TRAP_mock_rep1 <NA> <NA> <NA>
6 6 TRAP_mock_rep2 <NA> <NA> <NA>
7 7 TRAP_Sm_rep1 <NA> <NA> <NA>
8 8 TRAP_Sm_rep2 <NA> <NA> <NA>

So, I write:

>cor(m[, 7],m[, 8])

Is it right?

[blakeoft]

I have a couple questions for you. Are you following some sort of published analysis? Also, does each row of samples(cuff) have a column in m? In other words, what does

all(colnames(m) %in% samples(cuff)[, 2])

say when you enter it into your R session?

Edit: I realize that the code above isn't exactly what I meant to ask for. Can you paste what R prints for

colnames(m)

[Marcos Lancia]

all(colnames(m) %in% samples(cuff)[, 2])
[1] TRUE

I'm working by myself. I don't following any published analysis. Do you know any? The R help isn't good either.

[Marcos Lancia]

colnames(m)

[1] "SN16K_mock_rep2" "SN16K_mock_rep1" "SN16K_Sm_rep2" "SN16K_Sm_rep1"
[5] "TRAP_mock_rep1" "TRAP_mock_rep2" "TRAP_Sm_rep1" "TRAP_Sm_rep2"

[blakeoft]

I saw the prefix "TRAP", and it made me think of Trapnell, as in Cole Trapnell. This is why I was curious if you were working with some kind of sample data set or something.

Your code should be correct though,

cor(m[, 7], m[, 8])

should give you what you want. Have you tried checking the correlation between both of these columns with all of the others?

You might also compare the following values:

sum(m[, 7] == 0)
sum(m[, 8] == 0)
sum(m[, 7] == 0 & m[, 8] == 0)

to see if you have many more zeros in one the columns or if they don't share many of the same zeros.

My suggestions are shots in the dark, so I apologize if nothing enlightening happens.

[Marcos Lancia]

No, I´m not working with Cole Trapnell datasets, these data are mine.
Question: How can I be sure that data plotted are the same analyzed by correlation? Make some kind of matrix, maybe?
Thanks so much for writing, you've been very helpful.

[blakeoft]

After reading in some of my old cufflinks data, I've realized that my second post in this discussion has some incorrect code. Please plot these two figures for comparison:

csScatter(genes(cuff), "TRAP_Sm_rep1", "TRAP_Sm_rep2")
# compare to
plot(log(m[, 7] + 1), log(m[, 8] + 1))

These plots should look pretty similar. You should be able to tell that you're passing the right vectors into cor().

[Marcos Lancia]

Hi,
Your suggestion of plotting log(m[, ]+1) gaves me an idea. I made the cor(log(m[, ]+1)) and it worked! The cor() values are up to 0.95 all of them. Thanks for your help, mission accomplished, up to now.
Do you know how can I put labels in genes with differential expression? I tried with labels=T in scatterplots, but it didn´t work.