STAR RNA-seq Alignment

STAR (Spliced Transcripts Alignment to a Reference) is the gold standard for RNA-seq alignment — sensitive to novel junctions, fast, and the default in nf-core/rnaseq. The 2026 reality: STAR 2.7.11+ has improved memory management and 2-pass mapping is now standard for sensitive analyses. The cost: ~40 GB RAM for human genome index generation.

When to use

• Bulk RNA-seq with novel junction discovery (most cases).
• Single-cell RNA-seq (STARsolo).
• Chimeric / fusion detection (with --chimOutType).
• Long-read cDNA / ONT direct RNA (use --alignEndsType Extend...).

When NOT to use

• Genome with no annotation and small memory budget → use HISAT2 (this category) or minimap2 -ax splice.
• DNA-seq → use bwa-mem or bwa-mem2.
• Quantification-only → use pseudo-alignment (salmon / kallisto).

Prerequisites

• STAR ≥ 2.7.11
• samtools ≥ 1.19
• Reference FASTA + GTF
• ~40 GB RAM and 50 GB disk for human index

Core workflow

1. Generate the genome index with STAR --runMode genomeGenerate.
2. Align reads with STAR --runMode alignReads.
3. 2-pass mapping for sensitive novel junction discovery: first pass produces a SJ.out.tab, second pass uses it.
4. Sort and index the BAM (STAR can output sorted BAM directly with --outSAMtype BAM SortedByCoordinate).
5. Index the BAM with samtools index.

Code patterns

Generate the genome index (one-time, ~40 GB RAM for human)

STAR --runMode genomeGenerate \
    --runThreadN 16 \
    --genomeDir star_index/ \
    --genomeFastaFiles reference/genome.fa \
    --sjdbGTFfile reference/genes.gtf \
    --sjdbOverhang 149 \
    --genomeSAindexNbases 14

--sjdbOverhang 149 for 150 bp reads; for shorter reads, set to read_length - 1.

Paired-end alignment (most common)

STAR --runMode alignReads \
    --runThreadN 16 \
    --genomeDir star_index/ \
    --readFilesIn reads_R1.fq.gz reads_R2.fq.gz \
    --readFilesCommand zcat \
    --outFileNamePrefix sample1/ \
    --outSAMtype BAM SortedByCoordinate \
    --outSAMattrRGline ID:sample1 SM:sample1 PL:ILLUMINA LB:lib1 \
    --quantMode GeneCounts

Output:

• sample1/Aligned.sortedByCoord.out.bam — coordinate-sorted BAM
• sample1/ReadsPerGene.out.tab — gene-level counts (use this for DESeq2/edgeR)
• sample1/SJ.out.tab — splice junctions
• sample1/Log.final.out — alignment stats

2-pass mapping for novel junction discovery

# Pass 1: align and produce splice junctions
STAR --runMode alignReads ... --outFileNamePrefix pass1/
mv pass1/SJ.out.tab pass1_SJ.tab

# Pass 2: re-align with discovered junctions
STAR --runMode alignReads \
    --sjdbFileChrStartEnd pass1_SJ.tab \
    --genomeDir star_index_with_pass1_SJ/ \
    ... --outFileNamePrefix pass2/

Or use the simpler workflow: regenerate the index with the new SJ file and re-align.

Single-cell RNA-seq (STARsolo)

STAR --runMode alignReads \
    --genomeDir star_index/ \
    --readFilesIn sc_R1.fastq.gz sc_R2.fastq.gz \
    --soloType CB_UMI_Simple \
    --soloCBstart 1 --soloCBlen 16 \
    --soloUMIstart 17 --soloUMIlen 10 \
    --soloBarcodeReadLength 0 \
    --soloCellFilter EmptyDrops_CR \
    --outFileNamePrefix sc_outs/

Long-read cDNA / ONT

STAR --runMode alignReads \
    --genomeDir star_index/ \
    --readFilesIn long_reads.fq.gz \
    --outFilterMismatchNmax 5 \
    --outFilterMatchNmin 10 \
    --alignEndsType ExtendSoftClip

Output sorted BAM and skip the post-alignment sort

STAR ... --outSAMtype BAM SortedByCoordinate ...

This writes the BAM sorted and lets you skip samtools sort. You still need samtools index.

Multi-sample parallel alignment (shell loop)

for r1 in reads/*_R1.fq.gz; do"
  base=$(basename "$r1" _R1.fq.gz)
  r2="reads/${base}_R2.fq.gz"
  mkdir -p "star_out/${base}"
  STAR --runMode alignReads \
      --runThreadN 8 \
      --genomeDir star_index/ \
      --readFilesIn "$r1" "$r2" \
      --readFilesCommand zcat \
      --outFileNamePrefix "star_out/${base}/" \
      --outSAMtype BAM SortedByCoordinate \
      --outSAMattrRGline ID:${base} SM:${base} PL:ILLUMINA \
      --quantMode GeneCounts
done

Index the output BAM

samtools index sample1/Aligned.sortedByCoord.out.bam

Extract splice junctions for visualization

# Convert STAR SJ.out.tab to BED12 for IGV
awk 'BEGIN{OFS="\t"} $1 !~ /#/ {print $1, $2-1, $3, ".", $7, $4}' SJ.out.tab > sj.bed

Important flags

Common pitfalls

• Not enough RAM for human index. STAR --runMode genomeGenerate needs ~40 GB. STAR 2.7.11+ has --limitGenomeGenerateRAM to bound it, but at the cost of slower index generation.
• Forgetting --readFilesCommand zcat for .fq.gz inputs. STAR will try to read them as text and fail.
• Using a default --sjdbOverhang for non-150 bp reads. Set to read_length - 1 for best sensitivity.
• Ignoring 2-pass mapping. The first pass discovers novel junctions, the second pass is more sensitive. For novel transcript discovery, 2-pass is essential.
• STARsolo barcode config. CBstart + CBlen and UMIstart + UMIlen must match your library prep. Misconfiguration gives 0% barcodes-in-cells.

Validation

• samtools flagstat sample1/Aligned.sortedByCoord.out.bam — high mapping rate (≥85% for bulk RNA-seq).
• head sample1/Log.final.out — look for Uniquely mapped reads % ≥ 75%.
• awk '$4=="Uniquely mapped reads %"' Log.final.out should be ≥ 70% for most samples.
• ReadsPerGene.out.tab should have non-zero counts for known reference genes.

Open alternatives

References

• STAR paper: Dobin et al. 2013, Bioinformatics — 10.1093/bioinformatics/bts635
• STAR 2.7.11+ release notes: https://github.com/alexdobin/STAR/releases
• Companion: ors-bioinformatics-sequence-hisat2-alignment, ors-bioinformatics-omics-rna-seq-count-matrix-qc.

Changelog

• 1.0.0 (2026-06-10): Initial adaptation by Pradyumna Jayaram from bio-star-alignment (bioSkills-main/read-alignment/star-alignment).