Read-Based Variant Discovery

  • Read-based variant discovery is a computational approach used to identify genetic differences by analysing DNA sequencing reads and comparing them with a reference genome or another genomic sequence. Sequencing technologies generate millions to billions of short or long DNA reads, each representing a portion of the original genetic material. Bioinformatics algorithms examine these reads to identify nucleotide substitutions, insertions, deletions, and larger genomic rearrangements. Read-based variant discovery is a fundamental component of genome analysis and is widely applied in medical genetics, population genomics, evolutionary biology, cancer research, microbial genomics, and plant and animal breeding.
  • The main principle of read-based variant discovery is that sequencing reads from a sample contain evidence of its genetic sequence. When reads are aligned to a reference genome, differences between the observed bases and the reference sequence can indicate genetic variants. For example, if the reference genome contains an A at a particular position but many high-quality sequencing reads contain G, this may indicate a single-nucleotide variant. Similarly, gaps in read alignments may indicate insertions or deletions, while unusual read orientations, distances, or split alignments may indicate structural variants.
  • Read-based discovery is commonly used in reference-guided variant analysis. In this approach, sequencing reads are mapped to a reference genome, and computational tools identify positions where the sample differs from the reference. This differs from de novo variant discovery based on genome assembly, in which sequencing reads are assembled into longer sequences before comparison or analysis. Read-based methods are often computationally efficient and suitable for large datasets, but their accuracy depends on reference quality, read length, sequencing depth, mapping accuracy, and genome complexity.
  • The sequencing data used for read-based variant discovery can originate from several technologies. Short-read sequencing, particularly Illumina sequencing, produces relatively short DNA reads with generally high per-base accuracy and is widely used for SNV and small-indel discovery. Long-read technologies, including PacBio HiFi and Oxford Nanopore sequencing, generate longer reads that can span repetitive regions and complex genomic changes. Long reads are particularly useful for detecting larger insertions, deletions, inversions, translocations, and other structural variants. The choice of sequencing technology depends on the organism, genome characteristics, variant classes of interest, required accuracy, and available resources.
  • Read-based variant discovery usually begins with the preparation of sequencing libraries. Genomic DNA is extracted, fragmented or otherwise prepared, and processed to produce sequencing libraries. Depending on the technology, library preparation may involve adapter ligation, amplification, indexing, or other steps. Sequencing produces raw read files that contain nucleotide sequences and associated quality information. For short-read sequencing, the most common raw-data format is FASTQ, which stores each read together with its nucleotide sequence and quality scores.
  • Quality control is an essential first step because sequencing errors, adapter contamination, low-quality bases, and technical artifacts can produce incorrect variant calls. Tools such as FastQC and MultiQC are widely used to assess read quality, sequence composition, GC content, duplication levels, and adapter contamination. Preprocessing tools such as fastp and Cutadapt can remove adapters, trim low-quality bases, and filter unsuitable reads. For long-read sequencing, quality assessment may also consider read length, per-read accuracy, and technology-specific quality metrics. Preprocessing decisions should be documented because aggressive trimming or filtering can remove useful sequence information.
  • After quality control, sequencing reads are generally aligned to a suitable reference genome. Alignment software compares each read with the reference sequence and identifies its most likely genomic location. Short-read aligners such as BWA-MEM2 and Bowtie2 are widely used for DNA sequencing data, while minimap2 is commonly used for long-read alignment. The resulting alignments are stored in Sequence Alignment/Map (SAM), Binary Alignment/Map (BAM), or the compressed CRAM format. These files contain information about read positions, alignment quality, mismatches, insertions, deletions, and other alignment features.
  • Alignment quality strongly influences variant discovery. Reads that map to the wrong genomic location can generate false-positive variants, while reads that fail to align may contain genuine variants in highly divergent or repetitive regions. Mapping quality estimates confidence in the placement of a read, although its precise interpretation depends on the aligner. Multimapping reads, which align equally well or nearly equally well to several genomic locations, require special attention. Repetitive sequences, paralogous genes, segmental duplications, and reference gaps can make accurate mapping difficult.
  • Before variant calling, alignment files commonly undergo several processing steps. These may include sorting reads by genomic coordinate, creating alignment indexes, assigning read groups, marking duplicate reads, and checking alignment statistics. SAMtools is widely used for sorting, indexing, viewing, and summarizing SAM, BAM, and CRAM files. Picard tools can be used for tasks such as marking duplicates and collecting alignment metrics. Duplicate marking is particularly relevant for some short-read library types because PCR amplification can produce multiple copies of the same original DNA fragment. However, duplicate handling must be appropriate for the sequencing technology and experimental design; blindly removing reads can be unsuitable for some datasets, including certain amplicon and long-read workflows.
  • Variant calling is the process of identifying candidate genetic differences from sequencing reads and estimating the corresponding genotypes. A variant caller examines the aligned bases and evaluates evidence from the reads covering each genomic position. For an SNV, it considers the reference base, alternative bases, base quality, mapping quality, read depth, strand distribution, and other evidence. For an indel, it evaluates gaps and local sequence alignments. Some callers use probabilistic models, local haplotype assembly, Bayesian inference, or machine-learning methods to distinguish true variants from sequencing errors.
  • Different variant callers are designed for different data types and analytical purposes. GATK HaplotypeCaller uses local haplotype assembly to identify small variants, including SNVs and indels, from suitable sequencing data. FreeBayes uses a Bayesian framework and supports different ploidy settings. Bcftools provides tools for variant calling from aligned reads and supports individual and population-level analyses. DeepVariant uses deep-learning methods to predict variants from sequencing evidence. These callers may produce different results because they use different algorithms, assumptions, and evidence models. Their suitability should be assessed using the sequencing technology, genome, ploidy, sample design, and analysis objective.
  • A simplified read-based variant-discovery workflow can be represented as follows:

Raw sequencing reads (FASTQ)

              |

              v

       Quality control

              |

              v

    Read preprocessing

              |

              v

       Reference genome

              |

              v

       Read alignment

              |

              v

    SAM/BAM/CRAM processing

              |

              v

        Variant calling

              |

              v

      Variant quality control

              |

              v

     Filtering and normalization

              |

              v

       Variant annotation

              |

              v

 Biological interpretation

  • For SNV and small-indel discovery, variant callers commonly produce results in Variant Call Format (VCF) or its binary representation, BCF. VCF records describe the genomic position, reference allele, alternative allele, quality score, filtering status, and additional information. Sample genotype information can also be included. A simplified example is:

##fileformat=VCFv4.3

##reference=GRCh38

#CHROM  POS      ID  REF  ALT  QUAL  FILTER  INFO

chr1    123456   .   A    G    60    PASS    .

  • This record describes a substitution from A to G at position 123456 on chromosome 1 relative to the GRCh38 reference assembly. The quality and filtering fields provide information about the confidence assigned by the variant-calling workflow. A complete VCF file should include appropriate metadata declarations and relevant sample information.
  • Genotype information can be represented in the FORMAT and sample columns of a VCF file. For a typical diploid organism with a biallelic variant, 0/0 indicates two copies of the reference allele, 0/1 indicates a heterozygous genotype, and 1/1 indicates two copies of the alternative allele. A phased genotype may use a vertical bar, such as 0|1, to indicate that the alleles have been assigned to specific haplotypes. These genotype codes must be interpreted in relation to ploidy and the conventions of the variant file.
  • Read depth is one of the most important sources of evidence in read-based variant discovery. It describes the number of sequencing reads covering a genomic position or region. Higher depth generally provides more observations and can improve confidence in variant detection, although the relationship is not unlimited or uniform. Sequencing depth may vary across the genome because of GC content, library preparation, amplification, mapping difficulty, and other technical factors. Low coverage can cause genuine variants to be missed, while unusually high coverage may reflect duplicated regions, amplification bias, or mapping to multiple genomic copies.
  • Allele balance is another important measure, particularly for heterozygous variants. It describes the relative proportion of reads supporting the reference and alternative alleles. In a typical diploid sample with a heterozygous SNV, the alternative allele fraction may be close to 0.5, although deviations can occur because of sampling variation, mapping bias, copy-number changes, contamination, or biological mosaicism. In tumour samples, allele fractions can be affected by tumour purity, subclonal composition, and local copy number. Therefore, allele balance should be interpreted in the context of the sample and genomic region rather than treated as a fixed universal threshold.
  • Other variant quality measures include base quality, mapping quality, genotype quality, variant quality, strand bias, and read-position bias. Base quality estimates confidence in the observed nucleotide, while mapping quality reflects confidence in read placement. Genotype quality estimates confidence in a sample’s genotype call. Strand bias occurs when alternative alleles are supported disproportionately by reads from one sequencing direction. Read-position bias occurs when variant-supporting bases are concentrated near the ends of reads, where sequencing errors may be more common. These metrics help identify technical artifacts and improve variant-call reliability.
  • Variant filtering is performed to remove low-confidence calls and retain variants supported by sufficient evidence. Filtering may use minimum depth, genotype quality, mapping quality, allele balance, strand-bias statistics, and caller-specific quality measures. Some workflows use hard filters based on predefined thresholds, while others use statistical approaches such as variant quality score recalibration when the dataset and available training resources are appropriate. Filtering criteria should be selected according to the sequencing platform, organism, sample type, and study design. Overly strict filtering can remove genuine variants, while insufficient filtering can retain false positives.
  • Variant normalization is important when preparing variant calls for comparison, merging, and annotation. Small insertions and deletions may have multiple equivalent representations in repetitive DNA regions. Normalization commonly involves trimming redundant bases and shifting indels to a consistent position relative to the reference sequence. Tools such as bcftools norm can normalize variant records and check reference alleles. Before combining datasets, researchers should ensure that the reference genome assembly, chromosome naming conventions, variant representation, and filtering procedures are compatible.
  • Read-based discovery of structural variants requires additional evidence because large genomic changes may not be represented by a simple mismatch or short alignment gap. Several read-based strategies are used. Read-depth analysis identifies regions with increased or decreased sequencing coverage, which may indicate duplications or deletions. Paired-end mapping examines the orientation and distance between sequencing reads from the same DNA fragment. Discordant read pairs may indicate an insertion, deletion, inversion, or translocation. Split-read analysis identifies reads that align partly to different genomic locations, providing evidence for structural-variant breakpoints. These approaches may be combined to improve detection and characterization.
  • Structural-variant callers include Manta, DELLY, GRIDSS, and LUMPY for suitable short-read datasets. These tools use combinations of paired-end, split-read, and assembly-based evidence. For long-read data, Sniffles and cuteSV can identify insertions, deletions, and other structural variants. Read-depth methods for copy-number analysis include CNVkit, GATK gCNV, CNVnator, and Control-FREEC, depending on the data type and study design. The performance of each tool depends on coverage, read length, sequencing accuracy, genome complexity, and the class of structural variant being investigated.
  • Long-read sequencing has expanded read-based variant discovery because individual reads can span regions that are difficult to resolve with short reads. A long read may cover a repeat, an insertion, or a structural-variant breakpoint in a single molecule. PacBio HiFi reads combine relatively long sequence lengths with high consensus accuracy, while Oxford Nanopore sequencing can generate very long reads and supports direct sequencing of native DNA molecules. Long-read analysis still requires careful attention to platform-specific errors, alignment quality, coverage, and variant-calling methods.
  • Reference selection is a major factor in read-based variant discovery. Reads are generally aligned against a reference genome that provides the coordinate framework for identifying differences. If the reference is incomplete or genetically distant from the sample, reads may align poorly or fail to map. This can lead to reference bias, in which sequences more similar to the reference are preferentially aligned and detected. Reference bias can reduce the discovery of variants in divergent populations and complex genomic regions. Alternative reference genomes, haplotype-resolved assemblies, pangenomes, and graph-based references can help represent a broader range of genomic diversity.
  • Read-based variant discovery is widely used in medical genetics. Whole-genome and whole-exome sequencing can identify candidate variants associated with inherited disorders, while targeted sequencing can examine specific disease-related genes. Variant discovery is followed by annotation and interpretation using resources such as ClinVar, ClinGen, gnomAD, and Ensembl VEP. The presence of a variant does not automatically establish that it causes disease. Clinical interpretation requires consideration of population frequency, inheritance, phenotype, segregation, functional evidence, and the quality of the underlying data.
  • In cancer genomics, read-based discovery is used to identify somatic SNVs, indels, copy-number changes, and structural rearrangements. Comparing tumour sequencing data with an appropriate normal sample can help distinguish acquired somatic variants from inherited germline variation. Variant allele fractions and copy-number profiles may provide information about tumour heterogeneity and clonal evolution. However, tumour purity, subclonal variation, sequencing depth, and normal-cell contamination can complicate variant detection and interpretation.
  • In population genetics and evolutionary research, read-based variant discovery generates genome-wide datasets for studying genetic diversity, population structure, relatedness, migration, admixture, and natural selection. SNVs and indels are commonly used in genome-wide association studies and population analyses. Structural variants can provide additional information about genome evolution and adaptation. Population studies must account for sample relatedness, population structure, reference bias, and differences in sequencing depth to avoid misleading conclusions.
  • In agriculture and animal breeding, read-based variant discovery supports the identification of genetic differences associated with production, fertility, health, adaptation, and product quality. Whole-genome sequencing can identify candidate variants, while genotyping and genomic prediction methods can use selected markers across breeding populations. In livestock, read-based analysis can contribute to parentage verification, disease-variant screening, genomic selection, and diversity monitoring. In plants, it can help identify variants associated with yield, flowering time, stress tolerance, and disease resistance. Candidate associations should be validated in relevant populations before being used in practical breeding decisions.
  • Microbial genomics also relies on read-based variant discovery to investigate strain differences, pathogen evolution, antimicrobial resistance, and transmission patterns. Sequencing reads can be mapped to a reference strain or analysed against a broader set of genomes. Variant differences may be used to estimate relatedness among isolates, although the results depend on reference selection, recombination, genome content, and the quality of the sequencing data. For organisms with substantial gene-content variation, pangenome approaches may complement read-based mapping.
  • Several limitations affect read-based variant discovery. Mapping errors, sequencing errors, insufficient coverage, contamination, sample swaps, repetitive regions, and reference bias can all produce inaccurate results. Some variants may be missed when reads do not span the affected region or when the alternative sequence is absent from the reference genome. Short reads may not resolve complex rearrangements, while long-read datasets may require technology-specific error correction and specialized callers. Different software tools may also produce different variant calls because of differences in algorithms, parameters, and assumptions.
  • Reliable analysis therefore requires careful quality control and reproducible processing. Researchers should document the reference assembly, sequencing platform, read-processing procedures, aligner, variant caller, software versions, parameters, and filtering criteria. Raw sequencing data, alignment files, variant-call files, and quality-control reports should be retained where possible. Important variants may require independent validation using targeted sequencing, PCR, Sanger sequencing, or another appropriate method. The validation strategy should match the variant class and the biological question.
  • Read-based variant discovery is a foundational approach in modern bioinformatics because it converts sequencing reads into information about genetic differences. It combines read quality assessment, reference alignment, variant calling, quality filtering, normalization, and functional annotation. Its accuracy depends on the sequencing technology, reference representation, genome complexity, computational methods, and experimental design. Continued advances in long-read sequencing, graph genomes, machine learning, and haplotype-aware analysis are expanding the ability to identify and interpret genetic variation across individuals, populations, and species.

Author:

Leave a Reply

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