Mapping vs Alignment mode in Salmon

Question

[Sowmya]

Hello,

I am trying to understand the difference between alignment and mapping mode in salmon. I ran it on my bulk RNASeq samples in both modes, and compared the quant.sf files,

while the effective length seems to be associated with the length column in mapping mode,

Name                    Length  EffectiveLength     TPM     NumReads
ENST00000456328.2       1657    1407.000          0.000000        0.000
ENST00000450305.2       632     382.000           0.000000        0.000
ENST00000488147.1       1351    1101.000          1.534484        2.426
ENST00000619216.1       68      2.848             0.000000        0.000
ENST00000473358.1       712     462.000           0.000000        0.000
ENST00000469289.1       535     285.000           0.000000        0.000
ENST00000607096.1       138     4.663             0.000000        0.000
ENST00000417324.1       1187    937.000           0.000000        0.000
ENST00000461467.1       590     340.000           0.000000        0.000

it is not the case in alignment mode (it is a list of 100.000).

Name                  Length  EffectiveLength          TPM     NumReads
ENST00000456328.2       1657    100.000          0.818424        14.291
ENST00000450305.2       632     100.000          0.000000        0.000
ENST00000488147.1       1351    100.000          1.353094        23.627
ENST00000619216.1       68      100.000          0.000000        0.000
ENST00000473358.1       712     100.000          0.000000        0.000
ENST00000469289.1       535     100.000          0.000000        0.000
ENST00000607096.1       138     100.000          0.000000        0.000
ENST00000417324.1       1187    100.000          0.000000        0.000
ENST00000461467.1       590     100.000          0.000000        0.000

These are the commands that I used:

mapping:

salmon quant -i $salmon_index \
  --libType A -r $FASTQ/$sample.fastq.gz  \
  --validateMappings -p ${SLURM_CPUS_PER_TASK} \
  --dumpEq --useEM -o $OUTPUT_FOLDER

alignment:

salmon quant -t $REFERENCE_FASTA --libType U -a $BAM_FOLDER/$sample.bam \
  -o $OUTPUT_FOLDER --ont --noErrorModel -p 10

Any insights would be appreciated! Thanks!

Answer 1

[Rob]

This is because in alignment mode you have run with the --ont flag. The --ont flag turns off length correction (because ONT protocols don't perform fragmentation upstream of sequencing), and this is why you see an EffectiveLength of 100 for all transcripts. Also, as an aside, the --ont flag implements an error model explicitly designed for long-read alignments, so you shouldn't have to also specify --noErrorModel with this flag.

[Sowmya]

Thanks Rob!

That makes sense. Is there any consensus on what would be better here? Mapping mode or Alignment mode + the ONT flag?

[Rob]

If you are quantifying ONT data, then you should definitely use the alignment mode, using e.g. minimap2 to align against the transcriptome. Salmon's mapping mode isn't designed for long-read sequencing.

[Sowmya]

Thank you!