Split-Read, Discordant-Pair, and Depth-Based Methods

  • Split-read, discordant-pair, and depth-based methods are three major approaches used in bioinformatics to detect structural variants (SVs) from next-generation sequencing data. Structural variants include deletions, insertions, duplications, inversions, translocations, and copy-number variations (CNVs). These methods identify genomic rearrangements by examining different characteristics of sequencing reads aligned to a reference genome. Split-read methods use reads that align to multiple genomic locations, discordant-pair methods analyse unexpected relationships between paired-end reads, and depth-based methods investigate changes in sequencing coverage across genomic regions. Each approach provides different types of evidence, and their combination improves structural-variant detection and characterization.
  • Split-read analysis is based on identifying sequencing reads that span structural-variant breakpoints. A split read is a single sequencing read in which different portions align to separate positions in the reference genome. This occurs when the sequenced DNA contains a genomic rearrangement that is not represented in the reference sequence. For example, if a DNA fragment contains an insertion or deletion relative to the reference genome, a read spanning the breakpoint may align partially to one side of the altered region and partially to the other. The alignment discontinuity provides evidence of the structural variant and can help identify its approximate genomic location.
  • Split-read methods are particularly useful for identifying precise structural-variant breakpoints. For deletions, a split read may connect two genomic regions that are separated by a segment in the reference genome. For insertions, a read may contain additional sequence that is not present at the corresponding reference location. Inversions can be identified when portions of a read align in different orientations, while translocations may produce split alignments connecting different chromosomes. The accuracy of breakpoint identification depends on read length, sequencing quality, alignment accuracy, and the uniqueness of the genomic sequences surrounding the breakpoint.
  • One major advantage of split-read analysis is its ability to identify relatively small insertions and deletions that may not produce strong paired-end or read-depth signals. It can also provide nucleotide-level or near-nucleotide-level breakpoint information when sufficient sequence evidence is available. However, split-read detection becomes difficult in repetitive regions, segmental duplications, and genomic regions with low sequence complexity. Short reads may not contain enough unique sequence on both sides of a breakpoint to establish their correct genomic location. Sequencing errors and ambiguous alignments can also generate false-positive split-read signals.
  • Discordant-pair analysis, also known as paired-end mapping analysis, detects structural variants by examining the alignment relationships between two reads generated from the same DNA fragment. In paired-end sequencing, both ends of a DNA fragment are sequenced, and their expected orientation and approximate distance are determined by the library preparation method. In a typical paired-end library, the reads are expected to align in a particular orientation, commonly inward-facing, with an insert size falling within a defined distribution. When a structural variant is present, the distance, orientation, or chromosome location of the paired reads may differ from the expected pattern. Such read pairs are called discordant pairs.
  • Discordant-pair analysis can identify several types of structural variants. In a deletion relative to the reference genome, paired reads may align farther apart than expected because the corresponding DNA fragment lacks a segment present in the reference. In a tandem duplication, paired reads may show an abnormal orientation or mapping arrangement around the duplicated region. Inversions can produce read pairs with unexpected orientations, while translocations may be indicated by paired reads mapping to different chromosomes. Insertions may also generate abnormal paired-end distances when the inserted sequence is larger than the expected fragment length or differs from the reference genome.
  • The main advantage of discordant-pair analysis is that it can provide evidence for structural variants even when individual sequencing reads do not directly span their breakpoints. It is particularly useful for detecting medium-sized and large rearrangements and can identify candidate structural variants across the genome. However, its resolution is generally limited by the sequencing library’s insert-size distribution. It may identify a breakpoint region rather than its exact nucleotide position. Library preparation differences, unusual DNA fragment sizes, mapping errors, repetitive sequences, and sequencing artefacts can also produce discordant pairs that resemble genuine structural variants.
  • Depth-based analysis, commonly called read-depth analysis, detects structural variants by measuring the number of sequencing reads aligned to different genomic regions. Under relatively uniform sequencing conditions, regions with similar copy numbers are expected to have comparable read coverage after accounting for technical variation. A deletion generally produces reduced sequencing depth, whereas a duplication or amplification may produce increased read depth. By comparing observed coverage with expected coverage across the genome, bioinformatics tools can identify regions that may contain copy-number changes.
  • Read-depth analysis is particularly important for detecting copy-number variations (CNVs). For example, if a genomic region normally contains two copies in a diploid individual, a deletion affecting one chromosome may reduce the expected copy number to one. A duplication may increase the copy number to three or more. The observed sequencing coverage is used to estimate these differences. In population studies, read-depth patterns can also be compared among individuals to identify CNV polymorphisms and differences in genomic copy number.
  • Several statistical approaches are used in depth-based analysis. One common method divides the genome into fixed or variable-sized windows and calculates the number of reads aligned to each window. Coverage is then normalized to account for differences in sequencing depth, GC content, and other technical biases. Some methods use a hidden Markov model (HMM) to classify genomic regions into different copy-number states, such as deletion, normal copy number, or duplication. Other approaches use segmentation algorithms to identify genomic regions with consistently increased or decreased coverage. Normalization and statistical modelling are important because sequencing coverage is rarely perfectly uniform across the genome.
  • Depth-based methods have several advantages. They can identify large deletions and duplications even when their breakpoints are not directly spanned by individual reads. They are also useful for detecting copy-number changes across large genomic regions and can be applied to whole-genome sequencing and, with appropriate design, targeted sequencing or whole-exome sequencing. However, read-depth analysis generally provides less precise breakpoint information than split-read methods. Its accuracy depends on sequencing coverage, genomic window size, GC-content correction, mapping quality, sample ploidy, and reference-genome quality. Small structural variants may be difficult to detect, especially at low sequencing depth.
  • Several bioinformatics tools use these methods individually or in combination. Split-read and discordant-pair evidence is used by tools such as Manta, DELLY, LUMPY, and GRIDSS. Manta combines paired-end and split-read information to detect several classes of structural variants. DELLY uses paired-end and split-read evidence, while LUMPY integrates different types of alignment evidence within a probabilistic framework. GRIDSS uses local assembly and breakpoint information to characterize structural rearrangements. Depth-based CNV detection tools include CNVnator and Control-FREEC, which analyse read coverage and account for relevant technical factors. Other tools, such as GATK gCNV, provide methods for copy-number detection, particularly in appropriately designed sequencing datasets.
  • The general bioinformatics workflow begins with sequencing quality assessment. Raw FASTQ files are examined using FastQC and MultiQC to evaluate base quality, sequence composition, adapter contamination, and other quality indicators. Low-quality reads and adapter sequences may be removed using fastp or Cutadapt. Sequencing reads are then aligned to a reference genome using an appropriate aligner, such as BWA-MEM for short reads. Alignment files are commonly stored in BAM or CRAM format and processed using SAMtools and other supporting utilities. Mapping quality, read coverage, duplicate reads, and alignment characteristics are evaluated before structural-variant analysis.
  • Following alignment, different evidence types can be extracted from the sequencing data. Split-read analysis identifies reads with supplementary or multiple alignment segments, discordant-pair analysis identifies read pairs with unexpected orientation or insert sizes, and depth-based analysis measures coverage across genomic windows. Structural-variant callers examine these signals to generate candidate variants, estimate their genomic coordinates, and classify them according to their likely type. The results are often stored in Variant Call Format (VCF), which supports downstream filtering, annotation, comparison, and interpretation.
  • Combining split-read, discordant-pair, and depth-based methods can improve structural-variant detection because each approach compensates for limitations in the others. For example, a deletion may produce increased distances between paired-end reads, split reads spanning its breakpoints, and reduced sequencing depth across the deleted region. When all three types of evidence support the same event, confidence in the structural-variant call can increase. However, not every variant produces all three signals, particularly when the region is repetitive, the sequencing coverage is low, or the variant is complex. Therefore, the absence of one type of evidence does not necessarily mean that a candidate variant is incorrect.
  • After variant calling, candidate structural variants require filtering, comparison, and annotation. Quality filters may consider read support, mapping quality, breakpoint confidence, variant size, and consistency between different callers. Tools such as SURVIVOR can merge and compare structural-variant call sets, while Truvari can support benchmarking and comparison of variant calls. Annotation resources such as Ensembl, UCSC Genome Browser, AnnotSV, ClinVar, and gnomAD structural-variation resources can help identify affected genes, regulatory regions, and previously reported variants. Researchers may also investigate whether candidate variants overlap quantitative trait loci (QTLs), genome-wide association study (GWAS) signals, or known disease-associated genomic regions.
  • The biological interpretation of structural variants often requires integration with other genomic datasets. RNA sequencing (RNA-seq) can reveal whether a deletion or duplication affects gene expression, while DNA methylation and chromatin accessibility data can help identify changes in regulatory activity. Hi-C and other chromosome conformation capture technologies can investigate whether structural variants alter chromatin loops, topologically associating domains (TADs), or enhancer–promoter interactions. Combining structural-variant evidence with transcriptomics, epigenomics, and functional annotation can provide a more complete understanding of the consequences of genomic rearrangements.
  • In population genomics, these methods can be used to identify structural-variant differences among individuals, breeds, and species. Researchers can estimate variant frequencies, investigate population differentiation, and identify genomic regions potentially associated with adaptation or selection. Depth-based methods are particularly useful for studying CNV frequency differences, while split-read and discordant-pair approaches can identify breakpoint-level variation. Combining structural-variant analysis with SNP genotyping, haplotype analysis, and genome-wide association studies can improve understanding of the genetic architecture of complex traits.
  • In animal genetics and breeding, split-read, discordant-pair, and depth-based methods can support the discovery of variants associated with growth, milk production, meat quality, egg production, fertility, disease resistance, and environmental adaptation. Short-read sequencing provides an efficient approach for screening large animal populations, while long-read sequencing can be used to resolve complex variants and repetitive genomic regions. Accurate detection and interpretation of structural variants may improve genomic characterization and help identify candidate variants for further validation and genetic evaluation.
  • Despite their usefulness, these methods have several limitations. Repetitive DNA, segmental duplications, low sequencing coverage, GC-content bias, sequencing errors, and incomplete reference genomes can affect detection accuracy. Different tools may produce inconsistent variant calls because they use different algorithms, thresholds, and evidence combinations. Structural variants located in complex genomic regions may require long-read sequencing or targeted experimental validation. Techniques such as quantitative PCR (qPCR), digital droplet PCR (ddPCR), fluorescence in situ hybridization (FISH), and long-read sequencing can be used to validate selected candidate variants. Interpretation must also distinguish genuine biological effects from technical artefacts and population-specific background variation.
  • In modern bioinformatics, split-read, discordant-pair, and depth-based methods form the foundation of many short-read structural-variant detection pipelines. Split reads provide breakpoint evidence, discordant read pairs reveal abnormal relationships between paired sequences, and read-depth analysis identifies changes in genomic copy number. Their complementary use improves the detection and characterization of deletions, insertions, duplications, inversions, translocations, and CNVs. Integration with long-read sequencing, genome assembly, pangenomics, transcriptomics, and epigenomics provides increasingly comprehensive information about genome structure, genetic variation, disease mechanisms, evolution, and animal breeding.
Author:

Leave a Reply

Your email address will not be published. Required fields are marked *