Extended Help for Differential Expression Analysis Tools

Authors: orcid logoJennifer Hillman-Jackson avatar Jennifer Hillman-JacksonMelkeberhan Berhanu Degefa avatar Melkeberhan Berhanu DegefaAdd Contributions!

The error and usage help in this FAQ applies to most if not all Bioconductor tools.

  • DEseq2
  • Limma
  • edgeR
  • goseq
  • Diffbind
  • StringTie
  • Featurecounts
  • HTSeq-count
  • HTseq-clip
  • Kalisto
  • Salmon
  • Sailfish
  • DEXSeq
  • DEXSeq-count
  • IsoformSwitchAnalyzeR

Review your error messages and you’ll find some clues about what may be going wrong and what needs to be adjusted in your rerun. If you are getting a message from R, that usually means the underlying tool could not read in or understand your inputs. This can be a labeling problem (what was typed on the form) or a content problem (data within the files).

Expect odd errors or content problems if any of the usage requirements below are not met.

General

  • Are your reference genome, reference transcriptome, and reference annotation all based on the same genome assembly?
    • Check the identifiers in all inputs and adjust as needed.
    • These all may mean the same thing to a person but not to a computer or tool: chr1, Chr1, 1, chr1.1
  • Differential expression tools all require sample count replicates. Rationale from two of the DEseq tool authors.
    • At least two factor levels/groups/conditions with two samples each.
    • All must all contain unique content for valid scientific results.
  • Factor/Factor level names should only contain alphanumeric characters and optionally underscores.
    • Avoid starting these with a number and do not include spaces.
    • Galaxy may be able to normalize these values for you, but if you are getting an error: standardize the format yourself.
  • DEXSeq additionally requires that the first Condition is labeled as Condition.
  • If your count inputs have a header, the option Files have header? is set to Yes. If no headers, set to No.
    • If your files have more than one header line: keep the sample header line, remove all extra line(s).
  • Make sure that tool form settings match your annotation content or the tool cannot match up the inputs!
    • If you are counting by gene_id, your annotation should contain gene_id attributes (9th column)
    • If you are summarizing by exon, your annotation should contain exon features (3rd column)
  • Sometimes these tools do not understand transcript_id.N and gene_id.N notation (where N is a version number).
    • This notation could be in fasta or tabular inputs.
    • Try removing .N from all inputs, and check for the accidential creation of new duplicates!
  • Errors? Understanding the job log messages can be confusing! But are accessible and worth reviewing.
    • The good news is that usage in Galaxy produces the same error messages as direct usage.
    • This means that a search at the Bioconductor Support website can provide useful clues! Come back to the Galaxy Help forum with any remaining questions.

Remember, for any value in your inputs that is not a number, using only alphanumeric characters and optionally underscores _ with no spaces is what the authors recommend. Check your factor names, sample names, gene identifiers, transcript identifiers, and header lines in files.

Reference genome (fasta)

  • Can be a server reference genome (hosted index in the pull down menu) or a custom reference genome (fasta from the history).
  • Custom reference genomes must be formatted correctly.
  • If you are using Salmon or Kalisto, you probably don’t need a reference genome but a reference transcriptome instead!
  • More about understanding and working with large fasta datasets.

Reference transcriptome (fasta)

  • Fasta file containing assembled transcripts.
  • Unassembled short or long reads will not work as a substitute.
  • The transcript identifiers on the >seq fasta lines must exactly match the transcript_id values in your annotation or tabular mapping file.

Reference annotation (tabular, GTF, GFF3)

  • Reference annotation in GTF format works best.
  • If a GTF dataset is not available for your genome, a two-column tabular dataset containing transcript <tab> gene can be used instead with most of these tools.
  • HTseq-count requires GTF attributes. Featurecounts is an alternative tool choice.
  • Sometimes the tool gffread is used to transform GFF3 data to GTF.
  • DO use UCSC’s reference annotation (GTF) and reference transcriptome (fasta) data from their Downloads area.
    • These are a match for the UCSC genomes indexed at public Galaxy servers.
    • Links can be directly copy/pasted into the Upload tool.
    • Allow Galaxy to autodetect the datatype to produce an uncompressed dataset in your history ready to use with tools.
  • Avoid GTF data from the UCSC Table Browser: this leads to scientific problems. GTFs will have the same content populated for both the transcript_id and gene_id values. See the note at UCSC for more about why.
  • Still have problems? Try removing all GTF header lines with the tool Remove beginning of a file.
  • More about understanding and working with GTF/GFF/GFF3 reference annotation
Persistent URL
Resource PURL: https://gxy.io/GTN:F00008
Still have questions?
Gitter Chat Support
Galaxy Help Forum