WGS vs WES: Which to Sequence, and How to Analyze It

If you are choosing between whole-exome and whole-genome sequencing for yourself, pick 30x WGS. Exome sequencing targets roughly 1–2% of the genome (about 30–40 Mb of coding sequence), and it does so with a hybridization capture step that introduces GC-dependent coverage bias, uneven per-exon depth, and blind spots at the ends of capture probes. WGS gives you uniform coverage, better small-variant calling in the same coding regions the exome targets, structural variants, mitochondrial DNA at high depth, repeat expansions, and pharmacogenomic star alleles in genes like CYP2D6 that capture panels handle badly. The cost gap that once justified WES has mostly closed. Exome still wins in one scenario: you need very deep coverage of coding regions (somatic tumor work at 200–500x, or mosaicism hunting), where spending the reads on 1% of the genome is the whole point.

What the two assays physically are

WES: fragment DNA, ligate adapters, then hybridize the library to biotinylated oligo baits. The baits come from kits such as Twist Exome 2.0, IDT xGen Exome Hyb v2 and Agilent SureSelect. Then pull down with streptavidin beads, PCR amplify and sequence. Every step of that enrichment is a place where signal gets distorted. Baits do not tile perfectly, GC-rich first exons capture poorly, and the PCR after capture creates duplicates that eat into your effective depth. Different vendors define “the exome” differently, so a variant that is in-target for Twist may be off-target for SureSelect. Your callable region is a vendor artifact.

WGS: fragment, ligate, sequence. PCR-free library prep is standard now, which removes duplicate inflation and most of the GC bias. Coverage is close to Poisson around the mean, so a 30x average genome gives you ≥20x over roughly 95% of the callable genome, and the depth distribution over coding exons is more even than a 100x exome’s.

Two consequences matter for your analysis. First, WGS at matched cost typically detects more coding variants than WES, because WES loses regions entirely to capture failure. Second, WGS gives you variant classes that exomes cannot see. The classes are copy number and structural variants with breakpoint resolution, intronic splice-region variants, and deep intronic variants with known pathogenic effect. WGS also gives expansions in genes like FMR1 and RFC1, with the caveat that short reads resolve expansion presence better than exact size.

Files you should get, and what to check first

Insist on the raw data. The deliverable you want is FASTQ (or an unaligned CRAM), plus an aligned CRAM against a stated reference, plus a gVCF or VCF. Ask which reference: GRCh38 with the analysis set that masks false duplications, or T2T-CHM13. If someone hands you a GRCh37 VCF in 2026, you will spend a weekend on liftover and lose variants in the process.

Sanity checks before you interpret anything:

# depth and coverage uniformity
mosdepth --by 1000 --fast-mode sample sample.cram

# contamination and sex check
verifybamid2 --SVDPrefix resource/1000g.phase3.100k.b38.vcf.gz.dat \
  --Reference GRCh38.fa --BamFile sample.cram

# alignment summary
samtools stats sample.cram | grep ^SN

We look at the numbers below.

• mean autosomal depth (want ≥30x for WGS)
• fraction of genome at ≥20x (want ≥90–95%)
• FREEMIX contamination estimate (want <0.02, and anything above 0.05 means re-sequence)
• duplicate rate (PCR-free WGS should be under 5%)
• insert size mode (expect 350–450 bp)

For an exome, add on-target rate and the fold-80 base penalty from picard CollectHsMetrics: fold-80 above 2.0 means a fifth of your target is badly underserved.

The pipeline we would run

For germline small variants, the practical choice is DeepVariant plus GLnexus, or GATK HaplotypeCaller if you need the ecosystem.

# alignment (bwa-mem2 or dragmap)
bwa-mem2 mem -t 32 -R '@RG\tID:1\tSM:sample\tPL:ILLUMINA' \
  GRCh38.fa r1.fq.gz r2.fq.gz | samtools sort -@8 -o sample.bam

# small variants
run_deepvariant --model_type=WGS --ref=GRCh38.fa \
  --reads=sample.bam --output_vcf=sample.vcf.gz \
  --output_gvcf=sample.g.vcf.gz --num_shards=32

Use --model_type=WES with --regions targets.bed for an exome. Do not run the WGS model on exome data: the depth distribution is different and precision drops.

Structural variants need a second pass, and one caller is not enough. We run Manta (split reads and discordant pairs), plus a read-depth caller for CNVs, and require support from both before we believe anything:

configManta.py --bam sample.cram --referenceFasta GRCh38.fa --runDir manta/
manta/runWorkflow.py -j 16

Add ExpansionHunter for repeat loci with a supplied variant catalog, and a mitochondrial pass (the GATK Mitochondria pipeline realigns to a shifted reference so variants near the origin are not lost). For HLA and CYP2D6, short reads need dedicated tools: HLA-LA and Cyrius respectively. Generic callers will get these wrong and will not tell you.

Annotation: VEP with --cache --offline --everything, plus a gnomAD v4 allele-frequency annotation, ClinVar, and SpliceAI scores. Then filter. A 30x genome yields roughly 4–5 million variants against GRCh38, of which perhaps 20,000 are coding, a few hundred are rare and predicted loss-of-function, and a handful are in genes with real evidence. Frequency filters do most of the work: for a dominant condition, anything above 0.1% population frequency in gnomAD is almost never causal.

Interpretation is where this gets hard

Calling variants is a solved engineering problem. Deciding what a variant means is not, and the field has written down how to do it properly. Clinical WGS interpretation follows a structured process: phenotype-driven gene prioritization, ACMG/AMP evidence codes applied per variant, and explicit reporting of what was and was not assessed ¹. Population-scale sequencing has made clear how much of what you find is uncertain rather than actionable ².

Two cautions we take seriously. First, the ACMG-recommended secondary findings list exists because a small set of genes have both good evidence and clinical action attached. Most other variants of uncertain significance stay uncertain. Sequencing healthy people surfaces a large volume of ambiguous results, and the positive predictive value of an incidental finding in an unselected person is much lower than the same finding in someone with matching symptoms ³. Second, the professional guidance on population-level genome sequencing is deliberately cautious about unsolicited findings and about consent for data you have not yet decided to look at ⁴.

If a variant in your data looks clinically relevant, the next step is a clinician and an accredited diagnostic lab. It is not a deeper filter. Research-grade calls need orthogonal confirmation, and classification changes over time as evidence accumulates. Implementation reviews of clinical WGS consistently name interpretation capacity and variant reclassification, not sequencing, as the bottleneck ⁵.

Where exome still makes sense

Deep somatic sequencing. Tumor-normal panels and exomes at 200–500x detect subclonal variants at 5% allele fraction that a 30x genome cannot see. Mosaicism screening in blood or tissue, same reasoning. And retrospective work on an existing cohort where everyone else has WES, since cross-assay batch effects in coverage will wreck your joint calling otherwise.

For one person building a baseline they will re-query for a decade, the genome is the better substrate. It does not change, the FASTQs re-analyze cleanly against a new reference and a new caller, and the annotation databases improve underneath it.

Questions people also ask

What’s the difference between WGS and WES? WES sequences only protein-coding exons (roughly 1–2% of the genome) after a hybridization capture step. WGS sequences everything without capture. WES concentrates reads, WGS gives uniform coverage and access to intronic, structural, mitochondrial, and repeat variation.

Is WES or WGS better? WGS for a germline baseline in one person. WES for deep somatic or mosaicism work, or for matching an existing WES cohort. At current prices, the cost argument for WES in a single-person germline project has largely disappeared.

What are the disadvantages of whole exome sequencing? Uneven coverage from capture bias (GC-rich first exons are the usual casualties), a callable region that depends on which vendor’s baits were used, poor structural variant sensitivity, no deep intronic or regulatory variants, higher duplicate rates from post-capture PCR, and worse performance in segmental duplications and pharmacogenes.

What diseases can exome sequencing detect? Exome sequencing finds coding variants associated with Mendelian conditions across neurodevelopmental, metabolic, cardiac, and hereditary cancer predisposition genes, plus the ACMG secondary findings gene set. Detecting a variant is not a diagnosis. Classification and clinical correlation are a clinician’s job, with confirmation in an accredited lab.

How much does a WES test cost? Research-grade exomes run in the low hundreds of dollars as a raw assay. Clinical WES with interpretation and a signed report costs considerably more. The gap between the two is interpretation labor. Sequencing does not account for it.

Footnotes

1. Christina Austin‐Tse, Vaidehi Jobanputra, Denise Perry, et al. Best practices for the interpretation and reporting of clinical whole genome sequencing. npj Genomic Medicine, 2022. https://doi.org/10.1038/s41525-022-00295-z ↩

2. Tuuli Lappalainen, Alexandra J. Scott, Margot Brandt, et al. Genomic Analysis in the Age of Human Genome Sequencing. Cell, 2019. https://doi.org/10.1016/j.cell.2019.02.032 ↩

3. Noralane M. Lindor, Stephen N. Thibodeau, Wylie Burke. Whole-Genome Sequencing in Healthy People. Mayo Clinic Proceedings, 2017. https://doi.org/10.1016/j.mayocp.2016.10.019 ↩

4. Carla van El, Martina C. Cornel, Pascal Borry, et al. Whole-genome sequencing in health care. European Journal of Human Genetics, 2013. https://doi.org/10.1038/ejhg.2013.46 ↩

5. Petar Brlek, Luka Bulić, Matea Bračić, et al. Implementing Whole Genome Sequencing (WGS) in Clinical Practice: Advantages, Challenges, and Future Perspectives. Cells, 2024. https://doi.org/10.3390/cells13060504 ↩