Hi,

I am working with RNAseq data using EdgeR (steps below), while I was discussing some preliminary data analysis and observations, I cam across a question about the gene length. Does EdgeR trimmed mean of M values (TMM) account for gene length along with the sequencing depth and RNA composition?

While I was exploring more about this, I came across a couple of resources (links below):

• List item

EdgeR trimmed mean of M values (TMM) - accounts for sequencing depth, RNA composition, and gene length,

• List item

[A scaling normalization method for differential expression analysis of RNA-seq data: 2 It states that gene length is generally absorbed into a certain parameter and does not get used in the inference procedure. The focus of the TMM method is on estimating the relative RNA production of two samples, essentially a global fold change, by equating the overall expression levels of genes between samples under the assumption that the majority of them are not differentially expressed. Thus, while gene length biases are acknowledged as significant in gene expression analysis.

Sample metadata

#>   Samples Ind Event     Treatment
#> 1      S1    I1  5m Untreated
#> 2      S2    I1  9m   Treated
#> 3      S3    I2  5m Untreated
#> 4      S4    I2  9m   Treated
#> 5      S5    I3  5m Untreated
#> 6      S6    I3  9m   Treated

EdgeR Analysis

library(edgeR)
group.Treatment <-  factor(Sample_metadata$Treatment)
y <- DGEList(counts = gene_counts, group = group.Treatment, remove.zeros = TRUE)
keep <- filterByExpr(y)
y <- y[keep, , keep.lib.sizes=FALSE]
y <- calcNormFactors(y, method = "TMM")
logCPM = cpm(y, prior.count=1, log=TRUE)
Then, use logCPM values for downstream analysis such as to calculate fold changes per individual, plotting, and more...

Thank you,

Sabiha

TMM normalizes the library sizes, not the genewise read counts. Gene lengths are not relevant to the computation that TMM does.

In general, edgeR does not need to adjust for gene length in DE analyses because gene length cancels out of DE comparisons. Please see the section on normalization in the edgeR User's Guide.

"Gene length bias" as defined by Young et al (2010) is quite a different thing and is taken into account after the DE analysis when interpretting the results in terms of annotation categories. This type of gene length bias relates to the power to detect DE and is optionally adjusted for by goseq or by the edgeR methods goana() and kegga().

Reference

Young MD, Wakefield MJ, Smyth GK, Oshlack A (2010). Gene ontology analysis for RNA-seq: accounting for selection bias. Genome Biology 11, R14.