Whole Genome Sequencing

Whole genome sequencing (WGS) is the laboratory and computational determination of the nuclear and mitochondrial DNA sequence of an organism's genome in a single analytical workflow. Unlike targeted gene panels or exome sequencing, which interrogate only the ~1.5% of the human genome that encodes proteins (exons), whole genome sequencing captures the diploid 6.1-gigabase (haploid 3.055 Gb) human genetic architecture, including the ~98.5% non-coding regions composed of promoters, enhancers, introns, structural repeats, and the 16,569-bp mitochondrial circular chromosome.

In clinical and population genomics, human whole genome sequencing is routinely performed at a mean depth of coverage of 30× to 40×, requiring 90 to 120 gigabases of raw sequencing data per individual via short-read massively parallel sequencing (2 × 150 bp paired-end reads) or long-read single-molecule real-time sequencing (PacBio HiFi) and nanopore sequencing. At 30× depth, a single human genome yields 4.1 to 5.0 million single-nucleotide variants (SNVs), 700,000 to 800,000 small insertions and deletions (indels < 50 bp), and 20,000 to 27,000 structural variants (SVs ≥ 50 bp) relative to the GRCh38 or Telomere-to-Telomere (T2T-CHM13) reference assemblies.

Depth of Coverage, Lander-Waterman Statistics, and PCR-Free Library Preparation

The relationship between sequencing throughput and base-level coverage follows the Lander-Waterman Poisson model, where mean coverage depth C = (N × L) / G depends on read count N, read length L, and haploid genome size G (3.055 × 10⁹ bp). At a mean depth of C = 30×, the Poisson probability that any given base remains unsequenced is P(0) = e⁻³⁰ ≈ 9.3 × 10⁻¹⁴, ensuring that >99.5% of the mappable euchromatic genome is covered by at least 10 reads—the statistical threshold required to distinguish heterozygous alleles (50% expected allele fraction) from sequencing errors.

Clinical WGS workflows employ PCR-free library preparation. High-molecular-weight genomic DNA (0.5–1.0 μg) is sheared mechanically by focused acoustics (Covaris) or cleaved enzymatically to a 350–450 bp insert distribution, end-repaired, A-tailed, and ligated directly to indexed Y-shaped sequencing adapters without polymerase chain reaction amplification. Omitting PCR eliminates polymerase-induced GC-bias, preventing coverage dropout across GC-rich first exons (>70% GC content) and AT-rich intergenic regions, while keeping PCR duplicate rates below 2%.

Comparison of Whole Genome Sequencing Modalities and Exome Sequencing

Clinical Diagnostic Advantages Over Exome Sequencing

Whole genome sequencing resolves five classes of pathogenic variation that evade hybridization-capture exome sequencing. First, because PCR-free WGS avoids probe-hybridization capture kinetics, it provides uniform coverage across GC-rich coding exons where exome capture fails, rescuing ~1% to 3% of coding regions. Second, WGS detects distal intronic splice-altering mutations (located >50 bp from exon-intron boundaries) that activate cryptic splice sites and pseudoexons, as well as promoter, 5′/3′-UTR, and cis-regulatory enhancer mutations.

Third, uniform read depth across contiguous kilobases enables nucleotide-level breakpoint resolution of copy-number variants (CNVs), single-exon deletions, and balanced chromosomal rearrangements (inversions and reciprocal translocations) via split-read and discordant read-pair signatures. Fourth, haplotype-aware algorithms (ExpansionHunter) detect pathogenic short tandem repeat (STR) expansions, such as CAG trinucleotide expansions in HTT (Huntington disease) or CGG repeats in FMR1 (Fragile X syndrome). Fifth, off-target reads mapping to the high-copy mitochondrial genome yield 1,000× to 3,000× mtDNA coverage, enabling quantification of low-level mitochondrial heteroplasmy (>1% variant allele fraction).

Alignment, Base Quality Recalibration, and Variant Calling (GATK & DeepVariant)

Processing a 30× human genome through a bioinformatics pipeline begins with unaligned paired-end reads and Phred quality scores stored in compressed FASTQ files (~60–80 GB per genome). The Burrows-Wheeler Transform aligner BWA-MEM2 maps reads against the GRCh38 (with alternate decoy contigs) or T2T-CHM13 reference genome, sorting alignments and marking optical duplicates to generate a lossless CRAM file (~15–20 GB).

In the GATK Best Practices workflow, Base Quality Score Recalibration (BQSR) corrects systematic cycle- and context-dependent error covariates before GATK HaplotypeCaller performs local de novo De Bruijn graph assembly across active regions to call SNVs and indels into a Genomic Variant Call Format (gVCF) file. Alternatively, convolutional neural network callers such as DeepVariant convert read pileups into multi-channel image tensors (encoding base identity, quality, strand, and mapping quality), achieving >99.9% SNV F1-scores and >99.4% indel F1-scores in Genome in a Bottle (GIAB) benchmarks. Structural variants are concurrently genotyped using Manta, DELLY, or Parliament2 by integrating read-depth segmentation with split-read alignments.

Population Biobanks and Accelerated Neonatal Intensive Care WGS

Since the 1000 Genomes Project (2,504 genomes at low-pass coverage), population-scale WGS cohorts—including the UK Biobank (500,000 whole genomes at 30× depth), the All of Us Research Program, TOPMed, and the Genome Aggregation Database (gnomAD) (>76,000 whole genomes)—have cataloged hundreds of millions of rare human variants. These reference databases establish allele frequency filters (<0.1% minor allele frequency) and loss-of-function constraint metrics (LOEUF) for clinical variant interpretation.

In neonatal intensive care medicine, stat whole genome sequencing—pioneered by Stephen Kingsmore at Rady Children's Institute for Genomic Medicine and Euan Ashley at Stanford University—combines transposome-mediated library preparation, high-output flow cells or multi-flow-cell PromethION nanopore sequencing, and hardware-accelerated FPGA/GPU alignment (DRAGEN) to deliver a molecular diagnosis for critically ill newborns within 7 to 13.5 hours from blood draw.

Why is 30× mean coverage considered the clinical standard for germline whole genome sequencing, whereas somatic tumor WGS requires 80×–100× depth?

In constitutional (germline) genetics, every nucleated diploid cell carries either 0%, 50% (heterozygous), or 100% (homozygous) of a variant allele. Under binomial sampling atop a Poisson read-depth distribution at 30× mean coverage, >99% of bases receive at least 15 reads, providing sufficient statistical power to distinguish a 50% heterozygous allele from a 0.1% (Q30) sequencing error. Conversely, solid tumor biopsies contain admixtures of normal stromal cells and genetically heterogeneous cancer subclones; detecting a somatic mutation present at a 5% to 10% variant allele fraction requires 80× to 100× tumor WGS paired with a 30×–40× matched normal blood sample.

How much digital storage does a single 30× human whole genome require across FASTQ, BAM, CRAM, and VCF formats?

A 30× human genome (~100 gigabases of 2 × 150 bp reads) occupies 60 to 80 gigabytes (GB) as gzip-compressed raw FASTQ files. After alignment to GRCh38, the binary alignment map (BAM) file requires 80 to 120 GB. Converting the BAM to reference-based CRAM compression—which stores only nucleotide differences relative to the reference assembly and quantizes quality scores—reduces the alignment footprint by 60%–75% to 15 to 25 GB. Finally, the compressed VCF file containing only the ~5 million variant sites occupies 150 to 300 megabytes (MB), while an all-sites gVCF occupies 1.5 to 3 GB.

What is the difference between mapping WGS reads to GRCh38 versus the Telomere-to-Telomere T2T-CHM13 reference genome?

The GRCh38 human reference assembly contains ~151 megabases of unresolved gaps (N bases) across pericentromeric satellites, acrocentric short arms (chromosomes 13, 14, 15, 21, and 22), and segmental duplications, alongside collapse errors that generate false-positive variant calls in paralogous genes. Released in 2022, T2T-CHM13 added 200 megabases of finished sequence, closing all euchromatic and heterochromatic gaps. Mapping short- or long-read WGS data to T2T-CHM13 eliminates millions of spurious paralogous alignments, corrects 269 medically relevant gene models, and reduces false-positive SNV calls across previously misassembled segmental duplications by up to 12-fold.

1. Bentley, D. R., Balasubramanian, S., Swerdlow, H. P., et al. (2008). "Accurate Whole Human Genome Sequencing Using Reversible Terminator Chemistry." Nature, 456(7218), 53–59. https://doi.org/10.1038/nature07517
2. The 1000 Genomes Project Consortium. (2015). "A Global Reference for Human Genetic Variation." Nature, 526(7571), 68–74. https://doi.org/10.1038/nature15393
3. Poplin, R., Chang, P. C., Alexander, D., Schwartz, S., Colthurst, T., Ku, A., Newburger, D., Dijamco, J., Nguyen, N., Afshar, P. T., et al. (2018). "A Universal SNP and Small-Indel Variant Caller Using Deep Neural Networks." Nature Biotechnology, 36(10), 983–987. https://doi.org/10.1038/nbt.4235
4. Halldorsson, B. V., Eggertsson, H. P., Moore, K. H. S., Hauswedell, H., Eiriksson, O., Ulfarsson, M. O., Palsson, G., Hardarson, M. T., Oddsson, A., Jensson, B. O., et al. (2022). "The Sequences of 150,119 Genomes in the UK Biobank." Nature, 607(7920), 732–740. https://doi.org/10.1038/s41586-022-04965-x