Hi all,

I'm analyzing human RNA-seq data (rRNA-depleted total RNA, strand-specific, paired-end, ~150bp) and seeing a big gap between STAR and Salmon mapping rates. Would appreciate a sanity check.

Reference: GRCh38 primary assembly + Ensembl release 115 transcriptome (328,868 transcripts). Built a standard decoy-aware Salmon index (whole genome as decoy, k=31), following the standard tutorial approach.

Salmon quant command:

salmon quant \
  -i salmon_GRCh38_115_decoy_index \
  -l ISR \
  -1 SAMPLE-FP-trimmed.fq.gz \
  -2 SAMPLE-RP-trimmed.fq.gz \
  -p 8 \
  --validateMappings \
  --seqBias --gcBias --numBootstraps 30 \
  -o SAMPLE/salmon_quant

Results:

• STAR unique mapping (same genome/GTF): 93.05%
• Salmon mapping rate: 49.6% (sample A) and 61.7% (sample B), same dataset
• expected_format = ISR, compatible_fragment_ratio = 1.0 (so strandedness is not the issue)
• For sample A, ~14.7% of fragments mapped better to genomic decoys and ~15.3% were dropped for low alignment score

My hypothesis: since this is total RNA + rRNA depletion (not poly-A selection), a lot of intronic/pre-mRNA reads map fine to the genome (explaining STAR's 93%) but don't match annotated mature transcripts, so Salmon correctly discards them via the decoy.

Questions:

1. Is 50-65% mapping rate normal for rRNA-depleted total RNA with a decoy-aware index, or unusually low?
2. Is a ~10-point difference between two samples in the same dataset normal sample-to-sample variation, or a red flag?

Thanks!