Bio Read Alignment Hisat2 Alignment

Version Compatibility

Reference examples tested with: hisat2 2.2+, 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.

HISAT2 Alignment -- Graph-Indexed Spliced Mapping at a Quarter of STAR's Memory

"Align my RNA-seq reads with low memory" -> Map reads across exon-exon junctions with a hierarchical graph FM-index that fits a small machine -- because HISAT2 buys splice-aware alignment at ~7 GB instead of STAR's ~30 GB, its MAPQ is GATK-friendly, and its SNP-graph index can remove reference bias before a single read is mapped.

• CLI: hisat2 -p 8 -x index -1 R1.fq.gz -2 R2.fq.gz | samtools sort -@4 -o aligned.bam -

Scope: low-memory RNA splice-aware mapping with HISAT2 -- index building (plain / annotation-aware / SNP-graph), strandedness, the --dta transcript-assembly mode, and manual two-pass. Contig naming and the QC gate -> alignment-files. Feature-rich/high-RAM RNA alignment, native gene counts, and fusion detection -> star-alignment. Counting reads over genes -> rna-quantification. DE without a BAM -> rna-quantification/alignment-free-quant. OUT OF SCOPE: DNA (bwa-alignment/bowtie2-alignment), long reads (long-read-sequencing/long-read-alignment), HLA typing (HISAT-genotype, a separate tool).

The Single Most Important Modern Insight

1. The hierarchical graph FM-index is why HISAT2 exists: near-STAR spliced alignment at ~1/4 the RAM. HISAT2 uses one global FM-index to anchor a read plus ~55,000 small local graph FM-indexes (each ~56 kb), and extends a spliced read within the relevant local index rather than stitching genome-wide as STAR does. Most introns fit inside one local window, so spliced extension is a cheap local operation -- the resident human index is ~4-7 GB vs STAR's ~30 GB. That memory win is the reason to choose HISAT2; the cost is slightly lower novel-junction sensitivity than STAR two-pass and no native gene counts or fusion output.
2. The SNP/haplotype graph index removes reference bias in the index, and the MAPQ is GATK-friendly. A hisat2-build --snp --haplotype (or the prebuilt grch38_snp index) encodes millions of known variants as alternate graph nodes, so a read carrying a known alt allele traverses the alt node with no mismatch penalty -- the bias that over-counts the reference allele is removed structurally, for all those sites at once, without a per-sample personalized reference. (Private/novel variants still cause bias, so rigorous ASE still needs WASP or a personalized reference.) HISAT2 also assigns unique reads MAPQ 60 (not STAR's 255), so its output goes into GATK without the reassignment STAR needs.
3. --dta is for transcript assembly only, and using it for plain counting throws away reads. --dta raises the minimum anchor length required to report a de-novo spliced alignment, deliberately suppressing short-anchor junction reads -- because StringTie/Cufflinks cannot reliably assemble a transcript from a 3-5 bp anchor and such reads produce spurious isoforms. That trades junction sensitivity for assembly cleanliness, so --dta belongs only in a transcript-assembly pipeline; for plain gene counting it just discards usable junction reads. Strandedness (--rna-strandness RF for the common dUTP/TruSeq case) must also be set, or sense reads land in "no feature" and counts roughly halve.

How HISAT2 Splices (the mechanism in brief)

A read is seeded by the global FM-index, then the relevant ~56 kb local FM-index is selected and the read is extended across the junction within it: the unaligned remainder is anchored in the local index and extended by repeated FM-index extension. Because the spliced extension is a narrow, local operation rather than a genome-wide seed-cluster-stitch, HISAT2 needs far less RAM than STAR -- and evaluates a narrower set of candidate splice configurations, which is the source of both its speed/memory advantage and its slightly lower novel-junction sensitivity.

Tool Taxonomy

Decision Tree by Scenario

Default when uncertain: HISAT2 with --rna-strandness RF (verify the strand), streamed to a coordinate-sorted BAM; add --dta only for transcript assembly.

Build Index

# Plain genome-only index (cheap; supply junctions at align time with --known-splicesite-infile).
hisat2-build -p 8 reference.fa hisat2_index

# Annotation-aware (better short-anchor placement). NOTE: a full human --ss --exon build needs a LOT of RAM;
# prefer the prebuilt grch38_tran / grch38_snp_tran indexes, or pass junctions at align time instead.
hisat2_extract_splice_sites.py annotation.gtf > splice_sites.txt
hisat2_extract_exons.py        annotation.gtf > exons.txt
hisat2-build -p 8 --ss splice_sites.txt --exon exons.txt reference.fa hisat2_index

Basic Alignment with Strandedness

# RF = reverse-stranded (dUTP / Illumina TruSeq Stranded mRNA -- the common case). Verify, do not assume.
hisat2 -p 8 -x hisat2_index --rna-strandness RF \
    --rg-id sample1 --rg SM:sample1 --rg PL:ILLUMINA \
    -1 reads_1.fq.gz -2 reads_2.fq.gz \
    --new-summary --summary-file sample.summary.txt | \
    samtools sort -@ 4 -o aligned.sorted.bam -
samtools index aligned.sorted.bam
# Single-end stranded: --rna-strandness R (reverse) or F (forward). Unstranded: omit the flag.

For StringTie / Cufflinks (transcript assembly)

# --dta reports longer anchors the assemblers need; use ONLY for assembly, not for plain counting.
hisat2 -p 8 -x hisat2_index --rna-strandness RF --dta \
    -1 r1.fq.gz -2 r2.fq.gz | samtools sort -@ 4 -o aligned.bam -

Manual Two-Pass (cohort novel-junction discovery)

# Pass 1: discover novel junctions per sample.
for r1 in *_R1.fq.gz; do
    base=$(basename "$r1" _R1.fq.gz); r2=${r1/_R1/_R2}
    hisat2 -p 8 -x hisat2_index --novel-splicesite-outfile "${base}.novel.txt" \
        -1 "$r1" -2 "$r2" -S /dev/null
done
# Merge across the cohort so every sample sees the same junction set (avoids a per-sample junction batch effect).
cat *.novel.txt | sort -u > cohort.novel.txt
# Pass 2: re-align every sample with the shared novel-junction set.
for r1 in *_R1.fq.gz; do
    base=$(basename "$r1" _R1.fq.gz); r2=${r1/_R1/_R2}
    hisat2 -p 8 -x hisat2_index --rna-strandness RF --novel-splicesite-infile cohort.novel.txt \
        -1 "$r1" -2 "$r2" | samtools sort -@ 4 -o "${base}.bam" -
done