Question

Fishpond/Swish Interaction Test Error

0

Entering edit mode

Miguel ▴ 20

@52af32db

Last seen 10 months ago

Spain

Hello everyone,

I am currently working on an RNA-Seq data analysis. I am using Salmon + Fishpond/Swish. I have performed DE analysis at gene and transcript level. Now, I want to check if there is an interaction effect between the two variables of the project, but I get an error in my script. To give you some context, I have two variables (genotype and diet), which makes 4 experimental conditions appear. They are a total of 15 samples. They are not paired samples. Regarding the code, this is the part of my script to see the interaction:

y_g <- scaleInfReps(y_g)
y_g <- labelKeep(y_g)
set.seed(1)
y_g<-swish(y_g, x ="genotype", cov = "diet", interaction = TRUE)

Once I run it, the following error appears:

Error in infRepsArray[, cond2, ] : subscript out of bounds
sessionInfo( )

Could someone help me? If you need any extra information, please, tell me.

Thank you in advance!

fishpond swish • 1.0k views

ADD COMMENT • link 10 months ago Miguel ▴ 20

score 0 · Answer 1 · 2023-06-15

0

Entering edit mode

ATpoint ★ 4.1k

@atpoint-13662

Last seen 10 hours ago

Germany

Looks similar to Fishpond with unbalanced dataset -- the developer is working on a fix. Please see the GitHub issue in linked thread.

ADD COMMENT • link 10 months ago ATpoint ★ 4.1k

0

Entering edit mode

Yes, thanks for posting and this bug is now fixed on GitHub.

This is also now pushed to Bioc as version 2.6.1 (release) and 2.7.1 (devel), should be available in 24 hours.

ADD REPLY • link 10 months ago Michael Love 41k

1

Entering edit mode

Hi, Michael Love:

I have run and tested version 2.6.1. For my particular situation described above, the error was completely fixed. However, when I run a test with 14 samples (one sample less and distributed differently in the groups), the following error appears:

Error in sample.int(length(x), size, replace, prob) : 
cannot take a sample larger than the population when 'replace = FALSE'

Previously, when the script ran perfectly, the distribution of the 15 samples across the groups was:

         diet
genotype CT NR
      A  4  4
      B  3  4

Then, when the script crashed, the distribution of the 14 samples was:

        diet
genotype CT NR
      A  4  4
      B  3  3

PS: I ran in both situations the code indicated in the first entry of the post. Also, in both situations, the reference levels of genotype and diet are A and CT, respectively.

Hope you can help me. If you need any extra info, please, tell me. Thank you!

ADD REPLY • link 10 months ago Miguel ▴ 20

1

Entering edit mode

I'll take a look now.

ADD REPLY • link 10 months ago Michael Love 41k

1

Entering edit mode

Thanks for noting this.

This should be fixed in versions *.2, and you can check with:

install_github("mikelove/fishpond")

ADD REPLY • link 10 months ago Michael Love 41k

1

Entering edit mode

It works perfectly. Thank you so much.

ADD REPLY • link 10 months ago Miguel ▴ 20

0

Entering edit mode

Thank you both.

ADD REPLY • link 10 months ago Miguel ▴ 20