Merge pull request #1366 from pmoris/improve-docs-count-files

Clarify docs on different tximport count files

CHANGELOG.md

@@ -17,6 +17,7 @@ Special thanks to the following for their contributions to the release:
 - [Luke Zappia](https://github.com/lazappi)
 - [Matthias Zepper](https://github.com/MatthiasZepper)
 - [Maxime Garcia](https://github.com/maxulysse)
+- [Pieter Moris](https://github.com/pmoris)
 - [Rob Syme](https://github.com/robsyme)
 - [Thomas Danhorn](https://github.com/tdanhorn)
 
@@ -120,6 +121,7 @@ Thank you to everyone else that has contributed by reporting bugs, enhancements
 - [PR #1362](https://github.com/nf-core/rnaseq/pull/1362) - Move multiqc module prefix for nf-test to module
 - [PR #1363](https://github.com/nf-core/rnaseq/pull/1363) - Minor updates of nf-core modules and subworkflows
 - [PR #1363](https://github.com/nf-core/rnaseq/pull/1363) - Update dupradar script
+- [PR #1366](https://github.com/nf-core/rnaseq/pull/1366) - Clarify docs on different tximport count files
 - [PR #1367](https://github.com/nf-core/rnaseq/pull/1367) - Clarify design formula and blind dispersion estimation
 
 ### Parameters

docs/output.md

@@ -684,9 +684,10 @@ The principal output files are the same between Salmon and Kallsto:
   - `<pseudo_aligner>.merged.gene_counts.tsv`: Matrix of gene-level raw counts across all samples.
   - `<pseudo_aligner>.gene_tpm.tsv`: Matrix of gene-level TPM values across all samples.
   - `<pseudo_aligner>.gene_counts.rds`: RDS object that can be loaded in R that contains a [SummarizedExperiment](https://bioconductor.org/packages/release/bioc/html/SummarizedExperiment.html) container with the TPM (`abundance`), estimated counts (`counts`) and transcript length (`length`) in the assays slot for genes.
-  - `<pseudo_aligner>.merged.gene_counts_scaled.tsv`: Matrix of gene-level library size-scaled counts across all samples.
+  - `<pseudo_aligner>.merged.gene_lengths.tsv`: Matrix of average within-sample transcript lengths for each gene across all samples.
+  - `<pseudo_aligner>.merged.gene_counts_scaled.tsv`: Matrix of gene-level library size-scaled estimated counts across all samples.
   - `<pseudo_aligner>.merged.gene_counts_scaled.rds`: RDS object that can be loaded in R that contains a [SummarizedExperiment](https://bioconductor.org/packages/release/bioc/html/SummarizedExperiment.html) container with the TPM (`abundance`), estimated library size-scaled counts (`counts`) and transcript length (`length`) in the assays slot for genes.
-  - `<pseudo_aligner>.merged.gene_counts_length_scaled.tsv`: Matrix of gene-level length-scaled counts across all samples.
+  - `<pseudo_aligner>.merged.gene_counts_length_scaled.tsv`: Matrix of gene-level length-scaled estimated counts across all samples.
   - `<pseudo_aligner>.merged.gene_counts_length_scaled.rds`: RDS object that can be loaded in R that contains a [SummarizedExperiment](https://bioconductor.org/packages/release/bioc/html/SummarizedExperiment.html) container with the TPM (`abundance`), estimated length-scaled counts (`counts`) and transcript length (`length`) in the assays slot for genes.
   - `<pseudo_aligner>.merged.transcript_counts.tsv`: Matrix of isoform-level raw counts across all samples.
   - `<pseudo_aligner>.merged.transcript_tpm.tsv`: Matrix of isoform-level TPM values across all samples.
@@ -727,11 +728,11 @@ Transcripts with large inferential uncertainty won't be assigned the exact numbe
 
 The [tximport](https://bioconductor.org/packages/release/bioc/html/tximport.html) package is used in this pipeline to summarise the results generated by Salmon or Kallisto into matrices for use with downstream differential analysis packages. We use tximport with different options to summarize count and TPM quantifications at the gene- and transcript-level. Please see [#499](https://github.com/nf-core/rnaseq/issues/499) for discussion and links regarding which counts are suitable for different types of analysis.
 
-According to the `txtimport` documentation you can do one of the following:
+According to the [`txtimport` documentation](https://bioconductor.org/packages/release/bioc/vignettes/tximport/inst/doc/tximport.html#Downstream_DGE_in_Bioconductor) you can do one of the following:
 
-- Use bias corrected counts with an offset: import all the salmon files with `tximport` and then use `DESeq2` with `dds <- DESeqDataSetFromTximport(txi, sampleTable, ~condition)` to correct for changes to the average transcript length across samples.
-- Use bias corrected counts without an offset: load and use `salmon.merged.gene_counts_length_scaled.tsv` or `salmon.merged.gene_counts_scaled.tsv` directly as you would with a regular counts matrix.
-- Use bias uncorrected counts: load and use the `txi$counts` matrix (or `salmon.merged.gene_counts.tsv`) with `DESeq2`. This does not correct for potential differential isoform usage. Alternatively, if you have 3’ tagged RNA-seq data this is the most suitable method.
+- Use the original (bias-uncorrected) counts _with an offset_: import all the salmon `quant.sf` files with `tximport` and then use `DESeq2` with `dds <- DESeqDataSetFromTximport(txi, sampleTable, ~condition)` to automatically construct an object with gene-level offsets in-built, which can be used by DESeq2 to automatically account for effective gene length effects across conditions. It's also possible to combine the `.merged.gene_counts.tsv` and `.gene_lengths.tsv` matrices output by this workflow to make your own object the same way - which is what [the nf-core differentialabundance workflow does](https://nf-co.re/differentialabundance/1.5.0/docs/usage#outputs-from-nf-corernaseq-and-other-tximport-processed-results). See the [DESeq2 module](https://github.com/nf-core/modules/blob/c8f7f481bf2ccd8f9c7f2d499d94e74d4b9e23ab/modules/nf-core/deseq2/differential/templates/deseq_de.R#L323) for the implementation details.
+- Use bias-corrected counts _without an offset_: load and use `salmon.merged.gene_counts_length_scaled.tsv` or `salmon.merged.gene_counts_scaled.tsv` directly as you would with a regular gene-level counts matrix. These files were created using the `tximport` argument `countsFromAbundance="scaledTPM"` or `"lengthScaledTPM"` to scale counts to library size or to library size and average transcript length respectively.
+- Use the original (bias-uncorrected) counts _without an offset_: load and use the `txi$counts` matrix (or `salmon.merged.gene_counts.tsv`) with `DESeq2`. This is generally **not** recommended, since it does not correct for potential differential isoform usage (the offset). However, if you have 3’ tagged RNA-seq data, then this _is_ the most suitable method, because the counts do not exhibit any length bias.
 
 > **NB:** The default Salmon parameters and a k-mer size of 31 are used to create the index. As [documented here](https://salmon.readthedocs.io/en/latest/salmon.html#preparing-transcriptome-indices-mapping-based-mode) and [discussed here](https://github.com/COMBINE-lab/salmon/issues/482#issuecomment-583799668), a k-mer size off 31 works well with reads that are 75bp or longer.