Differentially gene expression analysis

Question

[Nai]

I have normalized count matrix which have negative values too but no raw count matrix. I would like to find deferentially expressed genes. I am using Deseq2. I am executing the following commands and error below: count_data <- as.matrix(read.csv(file='matrix.csv', header=TRUE, sep=';', row.names = 1)) info <- as.matrix(read.csv(file='sample.csv', header=TRUE, sep=';', row.names = 1)) ddg <- DESeqDataSetFromMatrix(count_data, info, ~condition)

Error in DESeqDataSet(se, design = design, ignoreRank) : some values in assay are negative

Other things I checked: any(count_data < 0 ) True

I need help to resolve this issue. As I did earlier was performed on raw count matrix. Now I took this count matrix from online expression matrix from database.

Answer 1

[ATpoint]

The problem is clear by the error message. If it is published data you can try to get rsw count matrix feom srrvices like BioJupies. Select the study and click through the analysis. In the end you can download tables with raw counts and metadata. It is not guaranteed that the data provided as matrix in GEO that you have are raw counts. Often it is already manipulated and unsuitable for mentioned tools. Alternatively check the recount package which also offers raw counts for a great many of studies.

[Nai]

May you guide me for recount package?

[ATpoint]

Whst is unclear after reading the documentation?

Answer 2

[Antonio R. Franco]

If I am not wrong, DESeq2 heavily indicate that you need to use raw counts, not normalized counts.

[Nai]

I agree. Is it any tool to do so?

[Victor J. Fernández]

You can't use tools that peform differential expression analysis based on negative binomial model (like DESeq2 or EdgeR).

I recommend you Limma (https://bioconductor.org/packages/release/bioc/html/limma.html)