# Short Read vs. Long Read Alignment Tools Short read alignment tools map millions of short fragments (roughly 50 to 300 base pairs) with high per-base accuracy and assume the read covers only a small slice of its true genomic origin, while long read alignment tools map relatively few reads that can span tens of kilobases, tolerate higher and more complex error patterns, and can align a single read across repeats, structural variants, and splice junctions. The practical rule of thumb is simple: if your question is about single-nucleotide variants, point mutations, or [gene expression](/blog/guides/gene-expression) counts from standard short-read data, start with a short-read aligner such as BWA-MEM or a splice-aware tool like STAR. If your question involves structural variants, repetitive regions, full-length transcripts, or genomes you plan to assemble, start with a long-read aligner such as minimap2. The harder question is not which tool is best, because no single best bioinformatics software package wins every job, but which read type and aligner combination answers your actual biological question at an acceptable cost. This guide walks through the difference between short read and long read alignment tools, the algorithmic reasons behind that difference, a side-by-side comparison, application-specific recommendations drawn from published benchmarks, and the hybrid workflows that increasingly combine both read types. ## Key Vocabulary Before the Comparison A few terms come up constantly in any bioinformatics tools comparison of aligners, so define them once here. - **Read**: the raw sequence output by a sequencer. One molecule of DNA or cDNA, read once or multiple times. - **Alignment (or mapping)**: placing a read at a position on a reference genome or transcriptome and describing the match, mismatches, insertions, and deletions in a CIGAR string (a compact notation such as `100M10I40M` meaning 100 bases matched, 10 inserted, 40 matched). - **Mapping quality (MAPQ)**: a per-alignment confidence score. Higher values indicate the aligner believes the placement is unique and correct. - **Seed-and-extend**: the dominant alignment strategy. The tool finds short exact or near-exact matches (seeds) between read and reference, then extends those seeds into full alignments, often with a Smith-Waterman style dynamic programming computation. - **FM-index**: a compressed full-text index of the reference that allows fast backtracking search. BWA and Bowtie are built on it. - **Hash-based index**: a lookup table of reference words (k-mers) pointing to reference positions. Tools like LAST, SSAHA2, and YAHA use this approach. - **Splice-aware alignment**: alignment that can split a read across an intron, essential for RNA-seq because mature transcripts contain introns removed relative to the genome. - **Split or chimeric alignment**: alignment of one read as two or more segments at distant genomic loci. This is how aligners represent [structural variant](/blog/guides/structural-variant) breakpoints and, in RNA-seq, splice junctions. - **Error profile**: the pattern of sequencing errors. Short reads produce mostly random substitutions at very low rates. Older long-read data carried error rates around 15 percent with frequent indels and homopolymer errors. Modern PacBio HiFi reads are highly accurate but still show elevated indel error relative to Illumina, and long reads in general carry more insertion and deletion noise [2]. ## Why the Algorithms Had to Diverge The difference between short read and long read alignment tools is rooted in mathematics and in the physics of the sequencers, not in developer preference. Short-read aligners face a combinatorial problem: with a 150 bp read against a 3 billion base reference, short exact seeds of 20 to 30 bases can occur many times by chance, so the tool must enumerate candidate positions efficiently and pick the best. FM-index based tools (BWA-MEM, Bowtie2) solve this with backtracking search. The payoff is extreme speed on massive read counts, and near-perfect discrimination for reads that are mostly error free. The tradeoff is that these tools assume low error and limited read length. Their gap penalties and seed strategies are tuned for a 100 to 150 bp read with one or two mistakes, not for a 20,000 bp read with hundreds. Long-read aligners face the opposite problem: a 20 kb read almost always contains errors spread along its length, so no long exact seed may exist. Instead, tools like minimap2 use shorter, sparser seeds (minimizers), chain many noisy seeds into a colinear chain, and only then run expensive alignment on the chained region. The long read itself provides the discrimination that short seeds cannot, because a 20 kb sequence is essentially unique even with 10 percent error. Long reads also frequently align as split mappings, one read spanning two distant loci, which is exactly the signal needed for structural variant detection [3]. Benchmark work on Nanopore data illustrates the practical consequences. A comparison of BLASR, GraphMap, LAST, minimap2, NGMLR, and the short-read aligner BWA-MEM on three experimental Nanopore datasets found large performance differences for LAST, consistently good quality and speed for minimap2, and only slight differences among BLASR, GraphMap, BWA-MEM, and NGMLR, with the authors concluding that tool choice should follow the experimental design and downstream analysis [4]. ## Side-by-Side Comparison Table | Criterion | Short read alignment tools | Long read alignment tools | |--|--|--| | Typical read length | 50 to 300 bp | 5 kb to over 100 kb | | Dominant error type | Low-rate substitutions | Indels, homopolymer errors, and clustered low-quality segments | | Index type | FM-index (BWA-MEM, Bowtie2) or dense hash tables | Minimizer/hash-based seed chains (minimap2, LAST), hash tables with breakpoint graphs (YAHA) | | Gap handling | Few small indels, short gaps | Many large indels, split/chimeric mappings across breakpoints [3] | | Throughput per run | Very high (billions of reads) | Lower read counts, larger individual records | | Memory and index size | Large reference index, fast per-read search | Smaller index, heavier per-read chaining and extension | | Repeat resolution | Poor, high multimapping in repeats | Strong, reads span repeats entirely | | Structural variants | Indirect (split reads, discordant pairs) | Direct (read spans the breakpoint) | | Splice junctions | Splice-aware tools (STAR, HISAT2) required | Minimap2 splice mode or transcriptome alignment [1][9] | | Representative tools | BWA-MEM, Bowtie2, STAR, HISAT2, SALT [5] | minimap2, NGMLR, GraphMap, BLASR, LAST, YAHA [10][16] | | Typical best use | SNP and small indel calling, gene counting, ChIP-seq peaks | Assembly, SV detection, phasing, full-length transcript sequencing | ## Short Read Alignment Tools on Their Own Terms Short-read aligners remain the workhorses of genomics for one reason: Illumina-style sequencing delivers enormous read volumes at very low per-base error, and for questions about single bases, that combination is hard to beat. **BWA-MEM** is the default choice for aligning Illumina DNA reads to a large reference genome. It uses seeded maximal exact matches, a banded Smith-Waterman extension, and produces high mapping accuracy that other projects still use as the reference point. Hardware work replicating BWA-MEM on an FPGA platform achieved over 99.90 percent agreement of final results with the standard software implementation, which tells you both how widely trusted the tool is and how much engineering effort goes into accelerating it [6]. **Bowtie2** offers a complementary FM-index design with flexible gap and mismatch penalties, and remains common for smaller genomes and resequencing projects. **STAR and HISAT2** are the splice-aware standard for RNA-seq. They index the genome with splice sites in mind so a read crossing an exon junction can be split across a large genomic gap. Where do short-read tools struggle, and what has the field done about it? - **Reference bias.** A linear reference represents one or a few individuals, so reads carrying population variants can misalign. SALT addressed this by aligning reads to a reference augmented with an SNP database, showing improved speed and accuracy over naive graph approaches while keeping memory costs manageable [5]. This is an active design axis for all aligners, short and long. - **Highly similar gene families.** KIR genotyping from short reads suffers cross-alignment between paralogs and outright alignment failure. The kir-mapper toolkit tackles this with specialized strategies for phasing alleles within and among KIR genes, validated against long-read data used as a truth set [7]. The lesson generalizes: when the region of interest is more polymorphic or more duplicated than average, short-read alignment accuracy degrades first, and either specialized pipelines or long reads are the fix. - **Read design interactions.** In ChIP-seq, a systematic evaluation of read length and pairing showed clear advantages for paired-end designs in alignment accuracy, peak resolution, and especially allele-specific binding detection [8]. This is a reminder that alignment quality is set partly at the experiment design stage, before any tool runs. So when to use short read alignment? When your data are short reads, your question lives at single-base resolution, and the regions you care about are not hopelessly repetitive. That covers most SNP and small indel calling, RNA-seq quantification, ChIP-seq peak calling, and targeted clinical panels. ## Long Read Alignment Tools on Their Own Terms Long-read aligners are built around three realities: reads are long enough to be self-unique, errors are frequent and structured, and one read may legitimately map in pieces. **minimap2** has become the de facto standard for long-read DNA alignment. Its minimizer-based seeding and chaining, followed by accurate extension, deliver speed and stable quality. In the Nanopore benchmark cited above, minimap2 achieved consistent alignment quality with good performance across datasets [4]. It also offers splice-aware presets for long RNA reads, though benchmarking shows this territory is more treacherous than DNA alignment (discussed below). **NGMLR and GraphMap** were designed with long-read error models in mind. NGMLR in particular uses a convex gap cost that helps it distinguish real large deletions from clusters of sequencing errors, which matters for structural variant calling. **LAST** uses adaptive scoring that adjusts to sequence complexity. It showed the largest differences from other tools in the Nanopore comparison, meaning its results depend strongly on dataset characteristics [4]. **BLASR**, developed for PacBio data, was an early long-read aligner and performed comparably to the others on Nanopore data, though the benchmark authors judged it the least applicable of the set to that platform [4]. **YAHA** deserves mention as a hash-based long-read-capable aligner with a distinctive feature: instead of reporting one or all alignments per read, it uses a directed acyclic graph to find the optimal set of alignments covering a query, with a biologically relevant breakpoint penalty. In testing it detected more breakpoints in less time than BWA-SW across structural variant classes and performed especially well on complex SVs with multiple breakpoints [3]. That makes it a useful example of the long-read design philosophy taken further: optimize for breakpoint discovery, not just for a single best placement. **BWA-SW family tools** show how long-read alignment scales in the cloud. A MapReduce-based implementation of BWA-SW on a custom parallel platform reported an average speedup of 6.7 over Bwasw-Cloud and reduced map phase makespan by about 30 percent, with Smith-Waterman optimization cutting execution time by 91.8 percent [9]. For laboratories with very large long-read volumes and cloud budgets, distributed alignment is a real option. ## Choosing by Application: Where Each Read Type Wins ![Decision path asking read length, then question type, leading to BWA-MEM, STAR, minimap2 or a long-read SV caller.](/article-images/illustrations/short-read-vs-long-read-alignment-tools-aligner-triage.webp) This is the section that answers "short read or long read alignment tools?" for the questions people actually have. ### SNP and small variant calling For single-nucleotide polymorphisms and short indels in non-repetitive DNA, short reads with BWA-MEM plus a standard caller remain the cheapest, most validated route. The analysis tooling around short-read variant calling is mature, and per-base accuracy from Illumina chemistry is excellent. Long reads are not excluded from this space. A benchmark of five long-read RNA-seq variant callers (Clair3-RNA, DeepVariant, longcallR, isoLASER, longcallR-nn) against GATK HaplotypeCaller on short-read RNA-seq, across ONT and PacBio library protocols and GIAB truth sets, found that sequencing characteristics were the strongest determinant of performance, and that upstream alignment strategy, including aligner choice and alignment transformation, measurably affected variant-calling accuracy [10]. The practical takeaway is that with long-read RNA data, alignment choices are not a neutral [preprocessing](/blog/data-analysis/what-is-data-preprocessing) step, they change your variant results. For laboratories that want both variant classes from one pipeline, integrated hybrid frameworks now exist. DNAscope Hybrid combines short- and long-read alignment and calling from the same sample, evaluated across HG002 to HG004 against GIAB v4.2.1 benchmarks, with the rationale that short reads bring base accuracy and cost efficiency while long reads resolve SVs and repeats, and the combination lowers the coverage needed from expensive long reads [2]. ### Structural variant detection This is where aligning short reads vs. long reads differs most sharply. Short reads detect SVs indirectly, through discordant pair distances, split reads, and read depth, and they struggle with insertions, repeats, and complex events. Long reads simply span the breakpoint. A systematic benchmark in pigs, sequencing five genetically diverse individuals with both short- and long-read platforms and evaluating 16 SV calling tools, found that long-read platforms detected many SVs missed by short reads with similar precision, that benchmark SVs of 200 to 500 bp (mostly insertions and deletions) had high validation rates, that the assembly-based caller SVIM-asm showed superior detection performance and resource efficiency, and that alignment-based tools performed well even at 5x depth [11]. In cancer genomics, characterizing somatic SVs has been limited by short-read technology. Work on the COLO829 melanoma reference line pursued the full spectrum of somatic SVs by aligning PacBio and ONT long-read data with Minimap2 and combining multiple alignment-based and assembly-based callers, motivated by evidence that sensitivity suffers when only a few alignment-based callers are used [12]. Two lessons follow. First, Minimap2 is the standard front end for long-read SV calling. Second, caller choice adds variance on top of aligner choice, so benchmark on your own data. ### RNA-seq and transcriptomics Splice-aware alignment of long error-prone reads is genuinely hard. An evaluation of RNA-seq splice-aware alignment tools, all originally built for short NGS reads but some claiming PacBio or ONT support, tested synthetic and real PacBio and MinION datasets and found that while some tools coped reasonably, others were unable to handle long error-prone reads, and error correction of long reads (self-correction or correction with external short reads) materially affected results [13]. If you are aligning long cDNA or direct RNA reads with a splice-aware tool, treat error correction as part of the alignment decision, not an afterthought. A newer strategy sidesteps genomic repeat and pseudogene problems entirely: align to the transcriptome first, then lift coordinates. T2G converts transcriptome-coordinate BAM files to genome coordinates while preserving CIGAR strings, tags, and per-base modification calls. Across nanopore direct RNA, [ribosome profiling](/knowledge/molecular-biology/ribosome-profiling), and paired-end Illumina data, transcriptome alignment plus T2G liftover gave higher on-target specificity and more accurate splice junctions than direct genome alignment, removed systematic misassignment of reads to processed pseudogenes (an 8.5-fold reduction in a reanalyzed RNA-binding protein experiment), and preserved isoform identity that genome alignment pools together [14]. Integration is also the story in circRNA discovery. Short-read RNA-seq detects backsplice junctions but cannot reliably reconstruct full circular sequences, so the CHRIS workflow used long reads enriched for full-length circRNAs as a guide for short-read alignment, rescuing circRNAs that short-read-only tools missed, and identified 6,445 non-canonical isoforms of known circRNAs in colorectal cancer samples [15]. Hybrid short-plus-long design here is not a compromise, it is the enabling method. ### Clinical diagnostics Diagnostic sequencing shows the tradeoff concretely. In a prospective trial of 110 individuals with cerebellar ataxia, short-read genome sequencing provided a genetic diagnosis for 38 percent of the cohort (40 of 110), including seven non-repeat-expansion pathogenic variants, while [long-read sequencing](/knowledge/bioinformatics/long-read-sequencing-technologies-pacbio-and-oxford-nanopore) with adaptive sampling targeted repeat expansion loci and identified pathogenic expansions in 29 individuals. The two methods were concordant on repeat expansions except for four SCA27B cases missed by long reads due to low read depth, and orthogonal testing revealed a 15 percent false positive rate for short reads versus 0 percent for long reads on that locus [16]. The technologies are complementary, with different failure modes, and manual review of repeat alignments improved outcomes for both. ### Metagenomic and pangenomic scale When your reference is not one genome but millions, index design dominates. XTree is a k-mer-based aligner that handles short and long reads against databases of up to millions of genomes, with pre-indexed resources including GTDB versions r214 through r226, fungal and protozoan representative genomes, and a 6.6 million genome viral compendium, reporting precision and recall at or above existing tools with more scalable indexing [17]. Long-read data benefit here because individual long reads resolve species and strain assignment that short fragments cannot, and the aligner must still keep the database tractable. ## Hybrid Workflows: Short Reads Fixing Long Reads and Vice Versa ![Worked calculation showing LSCplus needs about one third to one quarter of LSC run time and one twentieth of the error correction time.](/article-images/illustrations/short-read-vs-long-read-alignment-tools-lscplus-time-saved.webp) The most interesting development in recent years is that the two read types are increasingly used as mutual correction systems, and alignment is central to every step. **Long reads corrected by short reads.** Raw PacBio SMRT transcript data historically carried roughly 15 percent random error, making short-read error correction a standard preprocessing step. LSCplus improved on the LSC algorithm to require roughly one third to one quarter of the time and about one twentieth to one twenty-fifth of the error correction time, making hybrid transcriptome correction practical for ordinary laboratories [1]. **Choosing correction parameters automatically.** Error correction tools need manual tuning of parameters such as k-mer size, and the optimum is both tool- and dataset-dependent. Lerna builds a language model of uncorrected reads and uses a perplexity metric to find the k-mer size that maximizes alignment rate without needing a reference genome, automating what used to be a laborious grid search [18]. **Scrubbing without a reference.** For Nanopore reads where errors cluster into low-quality segments, MiniScrub uses read-to-read alignments from MiniMap, encodes them as images, and applies a convolutional neural network to predict and remove bad segments, producing assemblies with fewer mis-assemblies and large indel errors without needing a reference or high-ficiency short reads. **Short-read polishing that respects repeats.** Long-read-only bacterial assemblies typically retain homopolymer-length errors, and most short-read polishers fix these poorly in repeat regions because short-read alignment itself is unreliable there. Polypolish uses all-per-read alignments to repair errors in repeats that other polishers cannot, and performs best when combined with conventional short-read polishers [19]. Note the design insight: the polisher deliberately changes how short reads are aligned (using all alignments rather than one best alignment) because the alignment step was the bottleneck. **Integrated calling.** The DNAscope Hybrid pipeline mentioned earlier is the pipeline-level expression of the same idea, jointly aligning both read types so that each technology's error profile compensates for the other [2]. **Guided short-read alignment.** CHRIS reverses the direction of assistance, using long-read-resolved circRNA sequences as guides to rescue signal from short-read alignments [15]. The consistent pattern across these tools: the alignment step, not the read type, is where hybrid methods innovate. They change which alignments are computed (all versus best), which reads align to which reference (read-to-read versus read-to-genome), and which coordinate system the output uses. ## Practical Pipeline Considerations A few operational notes for actually running these tools. **Typical short-read commands.** For Illumina DNA reads against a human reference, the standard pattern is `bwa mem -t 8 reference.fa reads_R1.fq.gz reads_R2.fq.gz | samtools sort -o aligned.bam`. For RNA-seq, `STAR, genomeDir index, readFilesIn R1 R2` with genome generation via `STAR, runMode genomeGenerate`. These are standard usage patterns from the official documentation. **Typical long-read commands.** For Nanopore DNA, `minimap2 -ax map-ont reference.fa reads.fq.gz | samtools sort -o aligned.bam`. For PacBio HiFi, use the `map-hifi` preset. For long RNA reads, the `splice` or `splice:hq` presets. Choosing the wrong preset is one of the most common silent errors, because the presets encode different error models and splice handling. **Resource profile.** Short-read alignment is embarrassingly parallel across millions of tiny reads, so it scales horizontally on CPU clusters. Long-read alignment processes fewer but much larger records, with heavier per-record chaining and dynamic programming, so it is more sensitive to single-thread bottlenecks and memory configuration. Distributed options exist for large volumes. **Coordinate systems.** Decide early whether you need genome or transcriptome coordinates. As T2G demonstrates, transcriptome alignment followed by liftover can outperform direct genome alignment for splice accuracy and pseudogene discrimination, but downstream tools need genome coordinates, so the liftover step must preserve CIGAR strings and tags [14]. ## Common Mistakes and Limitations These are the errors that actually cost people weeks of reanalysis. 1. **Running long reads through a short-read aligner by default.** BWA-MEM on a 30 kb read either fails or produces fragmented, low-confidence mappings, because its seed and gap model assumes short, accurate reads. The Nanopore benchmark included BWA-MEM as a comparison and found it performed acceptably but without advantages over dedicated long-read tools [4]. Use a long-read aligner for long reads. 2. **Using the wrong minimap2 preset.** The presets encode error models and splicing behavior. Aligning Nanopore cDNA with a DNA preset drops splice junction awareness and silently misrepresents transcripts. 3. **Assuming alignment quality equals variant calling quality.** Caller choice adds substantial variance on top of the aligner, in both SV calling [12] and long-read RNA variant detection [10]. Benchmark aligner and caller combinations, not just aligners. 4. **Trusting a single SV caller.** Studies that combined a broader range of alignment-based and assembly-based callers found events that narrow tool panels missed [12]. 5. **Polishing with unreliable alignments in repeats.** Standard short-read polishers inherit the repeat-mapping weakness of short-read alignment. Polypolish exists specifically because one-best-alignment polishing fails in repeat sequences [19]. 6. **Ignoring error correction for long RNA alignment.** Splice-aware aligners built for short reads degrade on long error-prone reads, and correction, either self-correction or short-read assisted, changes outcomes [13]. 7. **Assuming long reads automatically mean better clinical answers.** The ataxia trial found long-read adaptive sampling missed four SCA27B cases due to low depth, and the short-read route carried a 15 percent false positive rate on that locus [16]. Depth and locus-specific error structure matter as much as technology choice. 8. **Treating mapping quality scores as comparable across tools.** MAPQ calibration differs between aligners. A MAPQ 60 from one tool is not numerically interchangeable with another, especially between short- and long-read tools with very different multimapping statistics. 9. **Forgetting that alignment happens in the context of read design.** Paired-end versus single-end, read length, and even capture design all shift alignment accuracy, as the ChIP-seq design evaluation showed [8]. One general limitation applies to all of this: benchmark results are reference- and organism-dependent, and a tool that wins on human data may not win on your organism, so any bioinformatics tools comparison should be replicated on a small subset of your own data before committing. ## Frequently Asked Questions ### Can I use BWA-MEM or Bowtie2 to align Nanopore or PacBio reads? You can, but you should not make it your default. These tools were designed around short, accurate reads, and their seed lengths and gap penalties penalize the dense indel error patterns of raw long reads. Dedicated long-read aligners such as minimap2 handle the error model natively and are faster at these read lengths. ### Which aligner should I start with for Oxford Nanopore data? Start with minimap2 and the preset matching your chemistry, then evaluate alternatives like NGMLR or LAST only if your benchmark on a data subset shows a problem. Independent comparisons found minimap2 delivered consistent alignment quality with good performance across Nanopore datasets, while other tools varied more with dataset characteristics. ### Do I still need short reads if I have long reads? Often yes, and increasingly the two are used together deliberately. Short reads polish long-read assemblies, correct long-read transcript errors, and provide cheap high-accuracy base calls, while long reads resolve the repeats and structural events that short reads cannot. Hybrid correction and polishing tools make this routine. ### Is there a single best bioinformatics software package for all alignment? No. The best tool depends on read length, error profile, genome complexity, and the downstream analysis. Any claim of a universal winner fails as soon as the application changes, which is why benchmarking on your own data subset is standard practice before committing to a pipeline. ### Why do long reads align better across repeats than short reads? Uniqueness comes from length. A 150 bp read may match hundreds of positions in a repetitive region, while a 20 kb read spans the entire repeat plus unique flanks, so even with substantial error it maps to one place. Specialized short-read methods can partially compensate by using all alignments rather than one best alignment, but they are working against a fundamental limit. ### Should I align RNA reads to the genome or the transcriptome? Both approaches are legitimate, and transcriptome-first alignment with a coordinate liftover step has shown higher on-target specificity and better splice junction accuracy than direct genome alignment in recent benchmarking, especially for discriminating processed pseudogenes. Choose based on whether your downstream tools need genome coordinates, and preserve tags and CIGAR strings if you lift over. ## Sources 1. [LSCplus: a fast solution for improving long read accuracy by short read alignment](https://pubmed.ncbi.nlm.nih.gov/27829364/) 2. [A novel and accelerated method for integrated alignment and variant calling from short and long reads.](https://pubmed.ncbi.nlm.nih.gov/41561440/) 3. [YAHA: fast and flexible long-read alignment with optimal breakpoint detection](https://pubmed.ncbi.nlm.nih.gov/22829624/) 4. [Comparative analysis of alignment tools for application on Nanopore sequencing data](https://doi.org/10.1515/cdbme-2021-2212) 5. [SALT: a fast, memory-efficient and SNP-aware short read alignment tool](https://doi.org/10.1109/BIBM47256.2019.8983162) 6. [High Accuracy Short Reads Alignment Using Multiple Hash Index Tables on FPGA Platform](https://doi.org/10.1109/ITOEC49072.2020.9141738) 7. [kir-mapper: A Toolkit for Killer-Cell Immunoglobulin-Like Receptor (KIR) Genotyping From Short-Read Second-Generation Sequencing Data.](https://pubmed.ncbi.nlm.nih.gov/40095784/) 8. [Systematic evaluation of the impact of ChIP-seq read designs on genome coverage, peak identification, and allele-specific binding detection](https://pubmed.ncbi.nlm.nih.gov/26908256/) 9. [Long Read Alignment with Parallel MapReduce Cloud Platform](https://pubmed.ncbi.nlm.nih.gov/26839887/) 10. [Systematic benchmarking of small variant calling pipelines for long-read RNA sequencing data](https://doi.org/10.64898/2026.04.29.721619) 11. [Systematic benchmarking of tools for structural variation detection using short- and long-read sequencing data in pigs](https://pubmed.ncbi.nlm.nih.gov/40060913/) 12. [Abstract 5056: Full spectrum of somatic structural variations (SVs) detection in COLO829 with long-read sequencing](https://doi.org/10.1158/1538-7445.am2025-5056) 13. [Evaluation of tools for long read RNA-seq splice-aware alignment](https://pubmed.ncbi.nlm.nih.gov/29069314/) 14. [T2G: fast and accurate transcriptome to genome BAM liftover for short and long read RNA sequencing](https://doi.org/10.64898/2026.10.20.760397) 15. [Integration of Short- and Long-Read RNA Sequencing Enables the Discovery of Circular RNAs](https://pubmed.ncbi.nlm.nih.gov/41329712/) 16. [A prospective trial comparing programmable targeted long-read sequencing and short-read genome sequencing for genetic diagnosis of cerebellar ataxia.](https://pubmed.ncbi.nlm.nih.gov/40015980/) 17. [XTree enables memory-efficient, accurate short and long sequence alignment to millions of genomes across the tree of life](https://doi.org/10.64898/2025.12.22.696015) 18. [Lerna: transformer architectures for configuring error correction tools for short- and long-read genome sequencing](https://pubmed.ncbi.nlm.nih.gov/34991450/) 19. [Polypolish: Short-read polishing of long-read bacterial genome assemblies](https://pubmed.ncbi.nlm.nih.gov/35073327/)