Hi,

I am interested in performing isoform analysis on short read data (150bp) using the edgeR package, following the example from the edgeR User's Guide, section "4.6 Differential transcript expression of human lung adenocarcinoma cell lines."

I ran my pipeline using the nf-core/rnasplice pipeline and obtained counts and TPM values: nf-core rnasplice output:

From the example 4.6 in the edgeR User's Guide, I tried importing the "quant.sf" files but experienced difficulties. Based on the example, I imported scaled counts as suggested:

# Define the path to your TSV file
file_path <- "/projects/salmon/tximport/salmon.merged.transcript_counts_scaled.tsv"

# Import the TSV file
scaled.counts <- read.delim(file_path, header = TRUE, sep = "\t", row.names = 1)

# Create DGEList object
y <- DGEList(counts = scaled.counts, samples = Samples_metadata)
dim(y)

Which of the following files makes the most sense to import for "4.6 Differential transcript expression of human lung adenocarcinoma cell lines"?

Counts from nf-core/rnasplice:

• salmon.merged.transcript_counts.tsv: Matrix of isoform-level raw counts across all samples.
• salmon.merged.transcript_counts_scaled.tsv: Matrix of isoform-level scaled raw counts across all samples.
• salmon.merged.transcript_counts_length_scaled.tsv: Matrix of isoform-level length-scaled raw counts across all samples.
• salmon.merged.transcript_counts_dtu_scaled.tsv: Matrix of isoform-level dtu scaled raw counts across all samples.

TPMs from nf-core/rnasplice:

• salmon.merged.transcript_tpm.tsv: Matrix of isoform-level TPM values across all samples.
• salmon.merged.transcript_tpm_scaled.tsv: Matrix of isoform-level scaled TPM values across all samples.
• salmon.merged.transcript_tpm_length_scaled.tsv: Matrix of isoform-level length-scaled TPM values across all samples.
• salmon.merged.transcript_tpm_dtu_scaled.tsv: Matrix of isoform-level dtu scaled TPM values across all samples.

You don't need any of these files. You need the catchSalmon function which reads quant.sf and the inferential (Bootstrap/Gibbs) replicates. Where in nf-core you can find that I cannot tell, but this is what the user guide tells, see also ?catchSalmon.

I haven't used nf-core but, from browsing through their documentation, they didn't seem to have any pipeline for differential transcript expression. To run the edgeR pipeline, you should run Salmon directly with --numGibbsSamples=50. Otherwise, edgeR has no way to measure the assignment uncertainty of each transcript count. See

Baldoni PL, Chen L, Smyth GK (2024). Faster and more accurate assessment of differential transcript expression with Gibbs sampling and edgeR v4. NAR Genomics and Bioinformatics 6(4), lqae151. https://doi.org/10.1093/nargab/lqae151

Note that the edgeR concept of "scaled counts" is quite different from scaled counts in nf-core.