Gene quantification

rustar-aligner can produce two flavours of quantification output during alignment, controlled by --quantMode. Both need a gene annotation: the tables of an index built with --sjdbGTFfile, or --sjdbGTFfile at alignment time. With neither, the run stops with STAR’s “could not open … geneInfo.tab” error.

Gene-level counts (GeneCounts)

Produces a <prefix>ReadsPerGene.out.tab file with one row per gene and four columns: gene ID, then read counts for unstranded, forward-stranded, and reverse-stranded protocols. Output is identical to STAR’s, so any downstream tool that already consumes STAR ReadsPerGene.out.tab works unchanged (e.g. DESeq2, edgeR, MultiQC).

rustar-aligner \
  --genomeDir /path/to/genome_index \
  --readFilesIn reads_1.fq.gz reads_2.fq.gz \
  --readFilesCommand zcat \
  --sjdbGTFfile gencode.v45.gtf \
  --quantMode GeneCounts \
  --outSAMtype BAM SortedByCoordinate \
  --outFileNamePrefix sample_

Output format

N_unmapped              <count>     <count>     <count>
N_multimapping          <count>     <count>     <count>
N_noFeature             <count>     <count>     <count>
N_ambiguous             <count>     <count>     <count>
ENSG00000000003.15      <count>     <count>     <count>
ENSG00000000005.6       <count>     <count>     <count>
...

The first four rows are summary categories (matching STAR / htseq-count semantics). Subsequent rows are per-gene counts. Pick the column that matches your library protocol:

Column	Strand assumption
2	unstranded
3	forward / htseq-count -s yes
4	reverse / htseq-count -s reverse

If you don’t know your library’s strandedness, all three columns help — you can compare them and infer.

Transcriptome-coordinate SAM (TranscriptomeSAM)

Produces a <prefix>Aligned.toTranscriptome.out.bam file with reads mapped to transcriptome coordinates (one record per matching transcript) instead of genome coordinates. This is the format consumed by RSEM and similar transcript-level quantifiers.

rustar-aligner \
  --genomeDir /path/to/genome_index \
  --readFilesIn reads_1.fq.gz reads_2.fq.gz \
  --readFilesCommand zcat \
  --sjdbGTFfile gencode.v45.gtf \
  --quantMode TranscriptomeSAM \
  --outSAMtype BAM SortedByCoordinate \
  --outFileNamePrefix sample_

Output filtering variants

--quantTranscriptomeSAMoutput controls what’s allowed in the transcriptome BAM:

Value	Meaning
BanSingleEnd_BanIndels_ExtendSoftclip	Default. RSEM-compatible: drop unpaired records, drop reads with indels, extend soft-clips into matches.
BanSingleEnd_ExtendSoftclip	Keep indels, still extend soft-clips.
BanSingleEnd	Keep indels and soft-clips as-is.

Pick the variant that matches your downstream tool’s expectations. RSEM defaults to BanSingleEnd_BanIndels_ExtendSoftclip.

Combining both

You can request both modes in the same run:

--quantMode GeneCounts TranscriptomeSAM

This emits ReadsPerGene.out.tab and Aligned.toTranscriptome.out.bam in addition to the normal alignment output.

Total RNA (ribo-depleted) libraries

For libraries with a large pre-mRNA content, --quantTranscriptomeUnspliced adds one unspliced <gene_id>-I target per gene to the transcriptome BAM (so pre-mRNA reads stop landing on retained-intron isoforms), --outSAMsplicingStatus Yes tags genomic alignments with their splicing status, and --quantGeneSplicing Yes counts spliced / unspliced / ambiguous reads per gene. All three are rustar-aligner extensions; see Bulk total RNA-seq.

Index-time vs alignment-time

For best speed, supply --sjdbGTFfile at --runMode genomeGenerate time and the transcript-level data structures get persisted into the genome directory. Then at alignment time you only need --quantMode TranscriptomeSAM (or GeneCounts, or STARsolo’s gene features); rustar-aligner reuses the persisted annotations.

If you supply --sjdbGTFfile at alignment time, its junctions are inserted into the genome and its transcript and gene info is used, even if the index has its own, as in STAR. The transcripts are put in STAR’s order, so the transcriptome BAM’s @SQ lines are the same as from an index built with that GTF. This is done on every run, so it adds startup time.

Strandedness summary

For --quantMode GeneCounts, all three strand columns are written regardless of library protocol — pick the right one downstream.

For --quantMode TranscriptomeSAM, the orientation is inferred from the transcript annotation in the GTF; no explicit strand parameter is needed.