carlosproca / cdnormbio Goto Github PK

Hi,

would you have any recommendations on how to handle non-converging data sets?

I tried to use normalize.svcd on a 3 data sets x 4 conditions matrix. It didn't converge. My naive wish were to increase the number of iterations. This isn't the way to do it though...

Are there some guidelines to be found?

I do not understand these two commands i.e. concept of adding offset.
Could you explain? Please..

offset.added <- rnorm( sample.n )
expr.data <- sweep( expr.data, 2, offset.added, "+" )

I have expression data of 3 sample. When i ran this command, i got following error.
Is replicates of sample is compulsory to run this command?

normalize.result <- normalize.svcd(expr.data, expr.condition,  stdvec.graph = "stdvec", p.value.graph = "p_value",  verbose = TRUE)
Error in normalize.svcd(expr.data, expr.condition, stdvec.graph = "stdvec",  : 
normalize.svcd: There must be 2 or more samples for each condition

expr.data.xlsx

Dear Dr. Roca,

Following your answers in the issues #3 and #4, indeed, I have 3 conditions with 3 replicates and 1 condition with 2 replicates (one was removed because of a bad reproducibility). So the size factors generated by SVCD have random variations.
In order to have more accurate data, I ran 50 times the normalization and I put the means into DESeq2, but it is not clear for me how to take the uncertainties into account. What could be the best way forward?

Thanks a lot for your answers.

Dear Dr Roca
I have expression data of multiple samples and conditions. When I ran this command:

normalize.svcd(expr.data.log2, expr.condition, verbose = T, convergence.threshold = c(0.01, 0.1, 0.01, 1)) # Default

I got following error.

between.condition.search.h0.feature
Error in rowMeans(edata[stdvec.feature, ]) :
'x' must be an array of at least two dimensions

I traced it back to normalize_svcd_aux.R, but I do not understand what is causing this error in there.

I pasted here an overview of the conditions just in case it is relevant.
Conditions:
table( expr.condition )
expr.condition
Ad_W0_CD Ad_W1_CD Ad_W1_HFD Ad_W12_CD Ad_W12_HFD Ad_W2_CD Ad_W2_HFD
6 6 6 6 6 6 6
Ad_W6_CD Ad_W6_HFD SVF_D3_CD SVF_D3_HFD SVF_W1_CD SVF_W1_HFD SVF_W12_CD
6 6 6 6 6 6 6
SVF_W12_HFD SVF_W2_CD SVF_W2_HFD SVF_W6_CD SVF_W6_HFD Ti_D3_HFD Ti_W1_HFD
6 6 6 6 6 2 2
Ti_W12_CD Ti_W12_HFD Ti_W2_HFD Ti_W6_CD Ti_W6_HFD Ti_WE_CD
2 2 2 2 2 3
Best regards

Dear Dr. Roca,
After reading your paper, I'm testing your normalization method in a RNA-seq experiment. I would like to have your opinion on my workflow:

1. normalize by SVCD on log2-scaled counts,
2. remove non expressed genes from the table,
3. estimate variance dispersion* by DESeq package,
4. apply negative binomial test** (the transformed normalized factors from SVCD are used as size factors),
5. remove the lowest expressed genes by an independent filtering based on the overall sum of normalized counts,
6. apply a 2nd test.

There are around 20000 transcripts from a microalgae in 4 different conditions to compare, with 3 replicates per condition.
Best regards

*https://www.rdocumentation.org/packages/DESeq/versions/1.24.0/topics/estimateDispersions
**https://www.rdocumentation.org/packages/DESeq/versions/1.24.0/topics/nbinomTest