Bio Read Alignment Star Alignment

Version Compatibility

Reference examples tested with: STAR 2.7.11+, samtools 1.19+

Before using code patterns, verify installed versions match. If versions differ:

• CLI: <tool> --version then <tool> --help to confirm flags

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

STAR RNA-seq Alignment -- The Junction Database, the 255 MAPQ, and the Strand Column Decide the Result

"Align my RNA-seq reads" -> Map reads across exon-exon junctions to the genome, building or reusing a splice-aware index from a GTF -- because an RNA read spans introns the genome does not contain, so the junction database (and its sjdbOverhang), the two-pass choice, the 255-unique MAPQ, and the strand column STAR reports are what actually determine the downstream counts and variant calls.

• CLI: STAR --runMode alignReads --genomeDir idx/ --readFilesIn R1.fq.gz R2.fq.gz --readFilesCommand zcat --outSAMtype BAM SortedByCoordinate

Scope: RNA splice-aware mapping with STAR -- index generation, two-pass, GeneCounts, chimeric/fusion output, and the MAPQ/strandedness traps. Contig-naming reconciliation and the QC gate -> alignment-files. Low-memory RNA alignment -> hisat2-alignment. Counting reads over genes/transcripts -> rna-quantification. Single-cell droplet/plate counting (STARsolo) -> single-cell. DE on known transcripts without a BAM -> rna-quantification/alignment-free-quant. OUT OF SCOPE: DNA (bwa-alignment/bowtie2-alignment), long reads (long-read-sequencing/long-read-alignment).

The Single Most Important Modern Insight

1. The splice-junction database is part of the index/run config, and a wrong sjdbOverhang or per-sample two-pass silently changes the answer. STAR builds short artificial junction-flank sequences from the GTF so reads with a short overhang on one exon can still be placed; --sjdbOverhang sets the flank length and should equal max(readlength) - 1 (default 100 is fine near 100 bp reads but degrades junction sensitivity for very short reads). Two-pass (--twopassMode Basic) discovers novel junctions and re-aligns, raising novel-junction sensitivity -- but it is PER-SAMPLE: each sample is re-aligned against its own augmented index, so a junction found only in the deep/disease sample is rescued asymmetrically, a batch effect that confounds junction/splicing/sQTL comparisons. The cohort-correct recipe is to pool every sample's pass-1 SJ.out.tab, filter, and feed one common --sjdbFileChrStartEnd to a uniform second pass. Per-sample two-pass is fine for plain gene-level DE; it bites splicing analyses.
2. STAR's MAPQ is 255-for-unique and a multiplicity code, so it breaks GATK and a copied MAPQ filter deletes every multimapper. Unique reads get MAPQ 255 -- the SAM "mapping quality unavailable" value -- which GATK treats as missing and drops, the classic silently-empty RNA VCF; fix it at align time with --outSAMmapqUnique 60. Multimappers get 3 / 1 / 0 for 2 / 3-4 / >=5 loci (pure locus count, no score information), so a generic samtools view -q 10 or featureCounts -Q 10 copied from a DNA pipeline DELETES every multimapper while keeping every unique -- a directional "discard paralog/gene-family/rRNA/pseudogene reads" filter that biases against recently-duplicated gene families. A hard MAPQ filter is almost never what an RNA analysis wants.
3. STAR reports library strandedness for free in GeneCounts, and getting strand wrong roughly halves counts. --quantMode GeneCounts emits a 4-column ReadsPerGene.out.tab: gene, unstranded, forward, reverse. Summing columns 3 and 4 reveals the protocol -- roughly equal is unstranded (use col 2), col 3 dominant is forward, col 4 dominant is reverse (the common dUTP/TruSeq case, use col 4). Never assume the strand: feeding the wrong column (or the wrong -s to a counter) sends sense reads to "no feature" and roughly halves the counts, distorting DE.

How STAR Places a Splice (the mechanism in brief)

STAR seeds with Maximal Mappable Prefix search on an uncompressed suffix array: it finds the longest exact prefix, and when that prefix ends (at a junction, mismatch, or read end) it restarts from the next base -- so a junction-spanning read naturally decomposes into seeds on two exons that STAR then stitches across the intron with an N (skipped-region) CIGAR. The stitch is scored with splice priors that penalize non-canonical motifs (--scoreGapNoncan -8), long introns (--scoreGenomicLengthLog2scale -0.25), and reward annotated junctions (--sjdbScore 2), plus a hard anchor floor (--alignSJoverhangMin): a novel junction on a tiny anchor carries almost no information, so it must clear both the soft score prior and the hard overhang minimum. This is why annotation (the sjdb) and two-pass matter for short-overhang and novel junctions.

Tool Taxonomy

Decision Tree by Scenario

Default when uncertain: STAR with --outSAMtype BAM SortedByCoordinate, --quantMode GeneCounts (to also read strandedness), and --twopassMode Basic for per-sample novel-junction work; set --outSAMmapqUnique 60 for any STAR -> GATK path.

Generate the Genome Index

# sjdbOverhang = max(readlength) - 1 (149 for 2x150 reads). Default 100 degrades junctions for short reads.
STAR --runMode genomeGenerate --runThreadN 8 \
    --genomeDir star_index/ \
    --genomeFastaFiles genome.fa \
    --sjdbGTFfile annotation.gtf \
    --sjdbOverhang 149
# Small genome (e.g. 5 Mb): add --genomeSAindexNbases <= min(14, log2(GenomeLength)/2 - 1), or STAR segfaults.

Basic Alignment

STAR --runThreadN 8 \
    --genomeDir star_index/ \
    --readFilesIn reads_1.fq.gz reads_2.fq.gz \
    --readFilesCommand zcat \
    --outFileNamePrefix sample_ \
    --outSAMtype BAM SortedByCoordinate
samtools index sample_Aligned.sortedByCoord.out.bam
# STAR already coordinate-sorts -- a subsequent `samtools sort` is redundant. Single-end: one file in --readFilesIn.

Two-Pass + Gene Counts + Strandedness

STAR --runThreadN 8 --genomeDir star_index/ \
    --readFilesIn r1.fq.gz r2.fq.gz --readFilesCommand zcat \
    --outFileNamePrefix sample_ \
    --outSAMtype BAM SortedByCoordinate \
    --twopassMode Basic \
    --quantMode GeneCounts \
    --outSAMattrRGline ID:sample1 SM:sample1 PL:ILLUMINA LB:lib1 \
    --outSAMmapqUnique 60        # so a downstream GATK RNA-variant step does not drop the 255 uniques
# STAR read groups use --outSAMattrRGline (SPACE-separated tags), NOT bwa's tab-delimited -R '@RG\t...'.
# GATK requires read groups; comma-with-spaces separates groups for multiple --readFilesIn files.

# Detect strandedness from ReadsPerGene.out.tab (skip the 4 N_* summary rows, sum cols 3 vs 4):
awk 'NR>4 {f+=$3; r+=$4} END {printf "fwd(col3)=%d  rev(col4)=%d -> use col %s\n", f, r, (f>2*r?"3 fwd": r>2*f?"4 rev":"2 unstranded")}' sample_ReadsPerGene.out.tab

ENCODE Long-RNA-seq Parameter Set

STAR --runThreadN 8 --genomeDir star_index/ --readFilesIn r1.fq.gz r2.fq.gz --readFilesCommand zcat \
    --outFileNamePrefix sample_ --outSAMtype BAM SortedByCoordinate \
    --outFilterType BySJout \
    --outFilterMultimapNmax 20 \
    --outFilterMismatchNmax 999 --outFilterMismatchNoverReadLmax 0.04 \
    --alignIntronMin 20 --alignIntronMax 1000000 --alignMatesGapMax 1000000 \
    --alignSJoverhangMin 8 --alignSJDBoverhangMin 1 \
    --sjdbScore 1 --outSAMattributes NH HI AS NM MD
# BySJout keeps only reads whose junctions passed the dataset-wide collapse; multimapNmax 20 retains real
# multi-locus genes; the 0.04 mismatch ratio scales with read length; intron caps at ~1 Mb cover human genes.

Fusion Detection

# Chimeric junctions for STAR-Fusion (params per the STAR-Fusion wiki).
STAR --runThreadN 8 --genomeDir star_index/ --readFilesIn r1.fq.gz r2.fq.gz --readFilesCommand zcat \
    --outFileNamePrefix sample_ --outSAMtype BAM SortedByCoordinate \
    --chimSegmentMin 12 --chimJunctionOverhangMin 12 --chimOutJunctionFormat 1 \
    --chimOutType Junctions
# Arriba instead reads chimeric alignments from the BAM: --chimSegmentMin 10 --chimOutType WithinBAM SoftClip.

Key Parameters (STAR defaults unless noted)

Per-Method Failure Modes

STAR 255 MAPQ into GATK

Trigger: STAR BAM (uniques at 255) fed to GATK RNA variant calling. Mechanism: GATK reads 255 as "MAPQ unavailable" and drops the read. Symptom: a silently empty or near-empty RNA VCF. Fix: --outSAMmapqUnique 60 at align time (then SplitNCigarReads in the GATK RNA workflow).