Direct comparison
HISAT2 vs STAR: Aligner Comparison
HISAT2 vs STAR: graph FM-index vs suffix-array indexing, RAM needed, alignment speed, and novel splice-junction sensitivity, per published benchmarks.
Written and maintained by CASRAI Editorial Board
Last updated
Ask CASRAI · included with Regulatory Radar
Ask about HISAT2 vs STAR: Aligner Comparison
Ask CASRAI answers research-administration questions and cites the passages behind every claim — and says so when the corpus does not cover something, instead of guessing. It comes with a Regulatory Radar subscription at $29 a month, alongside the daily digest of regulatory changes and the dashboard of what changed.
150 questions a day, on this site, over the API, or inside your own tools through the CASRAI MCP server.
Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays free to read, with no account and no card.
How do HISAT2, STAR compare side by side?
The table below compares HISAT2, STAR across 11 procurement-relevant dimensions, from indexing method through output format.
Side-by-side comparison
| Dimension | HISAT2 | STAR |
|---|---|---|
| Indexing method | Hierarchical Graph FM-index (HGFM): a Burrows-Wheeler-transform-based FM-index extended to represent the reference as a graph, so known variants (e.g. common SNPs) can be encoded directly into the index. | Uncompressed suffix arrays, the approach described in the original STAR paper (Dobin et al., 2013). Substring lookups are extremely fast, but the array itself must be built and held largely in memory. |
| Reference index size (human genome) | HISAT2's own documentation gives an index size of 6.2GB for the human genome plus 12.3 million common SNPs, with a 6.7GB memory footprint when running; a genome-only index (no SNPs) is smaller still. | STAR's documentation does not quote a fixed index size in GB, but ties memory need directly to genome size and annotation, which is why its RAM guidance (below) is stated as a floor rather than a fixed figure. |
| Minimum practical RAM for a human/mammal genome | Runs within roughly 6-8GB for a full human index with common variants included -- feasible on a well-specced laptop or a small cloud instance. | STAR's documentation states mammal genomes require "at least 16GB of RAM, ideally 32GB" for index generation and alignment -- effectively a server-class node or a memory-optimized cloud VM. |
| Alignment speed profile | Fast, and deliberately built to keep the speed of its FM-index-based predecessor (HISAT) while adding graph awareness; per-read alignment is efficient but the suffix-array approach below can edge it out in raw wall-clock time when memory is not a constraint. | Built explicitly around raw throughput: a widely cited RNA-seq aligner benchmarking study (Baruzzo et al., 2017, Nature Methods) describes STAR as released with "a RAM-intensive approach that dramatically increased speed" relative to earlier splice-aware aligners like TopHat2. |
| Sensitivity to novel (unannotated) splice junctions | The same Baruzzo et al. (2017) simulation-based benchmark found that "HISAT, HISAT2, and ContextMap2 are remarkably accurate even on the shortest anchors and without annotation" -- a direct measure of novel-junction detection strength. | The same benchmark places STAR among the most consistently accurate aligners for junction detection generally ("CLC, STAR, and NOVOALIGN" as top performers), without singling it out specifically for the short-anchor, no-annotation case the way HISAT2 is singled out. |
| Dependence on a supplied GTF annotation | Runs well with no annotation supplied at all; annotation can be added to bias alignment toward known transcripts but is not required for solid junction-finding. | Can run without an annotation, but STAR's documented two-pass mode -- which uses first-pass splice junctions to re-annotate the genome before a second alignment pass -- is the setting most likely to recover novel junctions, and it costs extra time and disk I/O. |
| Feasibility of running many samples in parallel on shared hardware | Its smaller memory footprint means more concurrent alignment jobs can fit on one shared cluster node or a modest number of cloud instances without hitting a RAM ceiling. | Each concurrent STAR job needs its own copy of the loaded genome index unless shared-memory mode is used, so running many samples in parallel on the same node needs proportionally more RAM -- a real practical constraint on shared HPC allocations. |
| Common companion pipeline | Pairs with StringTie (and formerly Ballgown) for transcript assembly and quantification -- the successor pipeline to the original TopHat/Cufflinks "Tuxedo" suite. | Underlies several widely used downstream tools that assume its output conventions, including STAR-Fusion for fusion-transcript detection and the alignment step in 10x Genomics Cell Ranger for single-cell RNA-seq. |
| Typical best-fit scenario | A lab with limited or shared compute memory, many samples to align in parallel, or cloud-cost sensitivity, and no hard dependency on a STAR-specific downstream tool. | A lab with a dedicated HPC node or memory-optimized cloud instance (32GB+), a need for the fastest possible per-sample alignment, or a pipeline (Cell Ranger, STAR-Fusion) that specifically expects STAR. |
| License | GNU General Public License v3 (GPLv3). | MIT License. |
| Output format | SAM, or BAM when piped through samtools; standard-compliant per the SAM/BAM specification. | SAM or BAM (STAR can output sorted BAM directly); standard-compliant per the SAM/BAM specification. |
Common questions
Common questions about HISAT2 vs STAR
Which aligner uses less memory, HISAT2 or STAR?
+
HISAT2, by a wide margin. HISAT2's graph FM-index keeps the full human genome index (with 12.3 million common SNPs included) at 6.2GB on disk and roughly 6.7GB in memory, per its own documentation. STAR's uncompressed suffix-array approach needs the documented minimum of 16GB of RAM for a mammal genome, and STAR's own guidance says 32GB is preferable. If available RAM is the binding constraint in your lab or on your shared cluster allocation, HISAT2 is the more forgiving choice.
Is STAR always faster than HISAT2?
+
Not universally, but STAR was explicitly designed around raw alignment throughput, trading memory for speed. Published benchmarking work (Baruzzo et al., 2017) frames STAR's 2013 release as a deliberately 'RAM-intensive approach that dramatically increased speed' relative to older splice-aware aligners. HISAT2 is also fast -- it inherited the speed of its FM-index-based predecessor, HISAT -- but where ample RAM is available and per-sample wall-clock time is the priority, STAR is generally the faster of the two.
Which aligner is more sensitive for detecting novel splice junctions?
+
Independent simulation-based benchmarking (Baruzzo et al., 2017, Nature Methods) found HISAT2 (alongside HISAT and ContextMap2) "remarkably accurate even on the shortest anchors and without annotation" -- conditions that specifically stress novel-junction detection. STAR was separately grouped among the most consistently accurate aligners for junction detection overall. Neither tool is clearly superior across all conditions; the practical takeaway from the literature is that both are strong choices for novel-junction discovery, and the memory/speed trade-off is a more reliable basis for choosing between them than a sensitivity gap.
Do I need a GTF annotation file to run HISAT2 or STAR?
+
No, both can align without one. An annotation is optional for HISAT2 and improves alignment to known transcripts when supplied. STAR can also run unannotated, but its documented two-pass mode -- which is the setting most likely to recover additional novel junctions -- uses first-pass splice-junction calls to build a refined, sample-specific annotation before the second alignment pass.
Can I realistically run STAR on a laptop?
+
For a human or other mammal genome, generally no. STAR's documentation states mammal genomes need at least 16GB of RAM for index generation and alignment, with 32GB recommended -- more than most laptops have available once the operating system and other processes are accounted for. HISAT2's smaller index (around 6.7GB in memory for a human genome with common SNPs) is realistic on a well-specced laptop. For small or non-mammalian genomes, STAR's memory requirement scales down accordingly.
Which aligner pairs better with downstream RNA-seq tools?
+
It depends on the pipeline you're already using. HISAT2 is the alignment step in the HISAT2-StringTie pipeline, the successor to the original TopHat-Cufflinks 'Tuxedo' workflow, for transcript assembly and quantification. STAR underlies tools that assume its specific output conventions, notably STAR-Fusion for fusion-transcript detection and the alignment stage inside 10x Genomics Cell Ranger for single-cell RNA-seq -- so if your downstream pipeline already expects STAR's output, that dependency usually settles the choice.
