Is Exome Sequencing Worth It, or Should You Just Sequence the Genome?

If a clinician is working up a suspected Mendelian disease, exome sequencing is a reasonable test to order and insurance sometimes pays for it. If you are buying sequencing for yourself, out of pocket, to have and analyze, buy a 30x whole genome instead. The price gap between a clinical-grade exome and a research-grade genome has closed to the point where the exome’s only remaining advantage is depth per dollar in the coding regions, and you pay for that with non-uniform coverage, capture artifacts, no reliable structural variant calling, and a dependence on whichever probe set the lab happened to use. Below is what the exome contains, how to audit one, and the specific failure modes worth knowing before you spend money.

What “the exome” means in practice

The exome is the protein-coding fraction of the genome: roughly 1-2% of 3.1 Gb, about 20,000 genes, and depending on the annotation you use, 30-60 Mb of sequence. That range matters. There is no single definition of the coding region. The commercial capture kits target different interval sets: IDT’s xGen Exome Hyb Panel v2 targets about 34 Mb, Twist’s Exome 2.0 about 36.8 Mb, Agilent SureSelect All Exon around 35 Mb of primary target with varying amounts of UTR padding. Each kit designs probes against a chosen annotation and drops regions that are too GC-rich, too repetitive, or too paralogous to capture cleanly. Work that built capture against the full GENCODE annotation found that standard exome kits leave a meaningful fraction of annotated coding exons untargeted or poorly covered¹.

So “whole exome” is a marketing term for “the subset of coding sequence a given vendor’s probes hybridize to.” When you get an exome, the first file you should ask for is the target BED. Without it you cannot tell a real homozygous reference call from a region with zero probes.

Where exome sequencing earns its keep

Gene discovery in rare disease. The original demonstration was a single patient sequenced with exome capture, where filtering against dbSNP and HapMap controls pointed to a homozygous missense variant in SLC26A3 and explained a congenital chloride diarrhea phenotype that had been misdiagnosed². That template — sequence a small number of affected individuals, filter common variants, look for rare protein-altering changes segregating with phenotype — produced hundreds of new disease-gene assignments over the following few years³. The method works because deleterious Mendelian alleles are heavily concentrated in coding sequence and because filtering against large population databases removes the overwhelming majority of candidates⁴.

Depth per dollar. At a fixed sequencing budget you can hit 100x mean on-target with an exome versus 30x on a genome. For somatic work, mosaicism, or any question where you need to call a variant at low allele fraction inside a known gene, that depth is real and useful.

Clinical interpretation infrastructure. Diagnostic labs have curated pipelines, trio analysis, and phenotype-driven prioritization built around exomes. Reported diagnostic yields in undiagnosed suspected-Mendelian cohorts generally land in the 25-30% range, which is high for a single test⁵. If you have a phenotype that suggests a monogenic cause, get a clinician to order this. Do not self-interpret a clinical exome.

Where it fails

Coverage uniformity. Hybridization capture is a stochastic, GC-biased process. Some targets pull down 200x, others 8x, in the same library. Whole-genome sequencing, even at lower mean depth, covers coding bases far more evenly, and a direct comparison found that WGS detects coding variants that WES misses, with the gap driven by capture dropouts rather than by depth⁶. That work also found a substantial number of coding variants called only in WGS, including ones in genes with established clinical relevance⁶.

Copy number and structure. You can force CNV calls out of exome depth data using GATK gCNV, ExomeDepth, or CODEX2, but they are relative-depth inferences across a panel of samples, they need dozens of matched-batch controls, and their resolution stops at roughly the single-exon level. Breakpoints, balanced translocations, inversions, and repeat expansions are invisible. Mitochondrial coverage depends on off-target reads and is inconsistent across kits.

Everything non-coding. Splice variants more than ~10-20 bp into an intron, promoter and enhancer variants, deep intronic cryptic splice sites, and most pharmacogenomic structural alleles (CYP2D6 in particular) are outside the target. Regulatory variation is where much complex-trait signal sits, and the exome discards it by design⁷.

Interpretation, not sequencing, is the bottleneck. A typical exome yields 20,000-25,000 variants against GRCh38, of which a few hundred are rare and protein-altering, and of those a handful will be genuinely interpretable. Most rare missense variants come back as variants of uncertain significance. Filtering strategy, control-set choice, and population ancestry matching all bias the result, and these are well-documented problems rather than solved ones⁸.

Auditing an exome you already have

Ask the lab for FASTQ or an aligned CRAM plus the target BED. If all they will give you is a PDF report, treat that as a red flag about whether you own the data.

Alignment and calling, roughly what we run:

bwa-mem2 mem -t 16 -R '@RG\tID:1\tSM:sample\tPL:ILLUMINA' \
  GRCh38_full_analysis_set.fa r1.fq.gz r2.fq.gz \
  | samtools sort -@8 -o sample.bam
gatk MarkDuplicates -I sample.bam -O sample.md.bam -M dup.txt
gatk HaplotypeCaller -R GRCh38.fa -I sample.md.bam \
  -L targets.bed --interval-padding 100 -ERC GVCF -O sample.g.vcf.gz

The --interval-padding 100 matters. Reads spill past probe boundaries, and calling only inside the strict target throws away usable flanking sequence including many canonical splice sites.

Quality metrics worth pulling before anything else:

picard CollectHsMetrics I=sample.md.bam O=hs.txt \
  R=GRCh38.fa BAIT_INTERVALS=baits.list TARGET_INTERVALS=targets.list
mosdepth --by targets.bed --thresholds 1,10,20,30 sample sample.md.bam

What we look at: PCT_TARGET_BASES_20X should be above 0.90, ideally 0.95. FOLD_80_BASE_PENALTY under 2.0 indicates decent uniformity. A value above 2.5 means you are wasting a lot of reads. PCT_OFF_BAIT between 0.25 and 0.40 is normal for hybrid capture. PERCENT_DUPLICATION above 0.20 on a PCR-free-ish prep suggests a low-complexity library. Then take the per-target output from mosdepth and list every gene you care about with a 20x fraction under 0.95. Those are the genes where a negative result means nothing.

For the VCF itself, bcftools stats gives you the Ti/Tv ratio. Exomes should sit near 3.0-3.3 overall. A value near 2.0 means you are carrying a lot of false positives. Het/hom ratio far from the expected ~1.5-2.0 for the sample’s ancestry points at contamination, which you can check directly with VerifyBamID2.

Annotation, and where to stop

Annotate with Ensembl VEP and not with a web uploader:

vep -i sample.vcf.gz --cache --offline --assembly GRCh38 \
  --everything --pick_allele_gene --plugin CADD --plugin SpliceAI \
  --custom gnomad.genomes.v4.vcf.gz,gnomADg,vcf,exact,0,AF \
  --vcf -o sample.vep.vcf.gz

Then filter to gnomAD AF below 0.001, consequence in the LoF or missense set, and intersect with ClinVar records reviewed by an expert panel or multiple submitters. A separate, narrower pass over the ACMG secondary-findings gene list (v3.2, 81 genes) is the only part of a personal exome with a well-defined action path. That path runs through a genetic counselor and a CLIA-confirmed retest. Research-grade calls are not diagnostic. If something appears in that list, confirm it in a clinical lab before anyone draws a conclusion from it.

Our view: spend the money on a 30x PCR-free genome with retained CRAM, then run the same coding analysis on it plus everything the exome could not see. If you are pursuing a specific undiagnosed phenotype, the right move is a clinician-ordered trio exome, because parental samples do more for interpretation than any amount of extra depth.

Questions people also ask

How much does an exome test cost? Clinical whole exome sequencing in the US typically lists between $1,500 and $5,000 depending on whether it includes trio analysis and CNV calling. Research or direct-to-consumer exomes with raw data delivery run a few hundred dollars. The sequencing is the cheap part, interpretation and reporting are what you are paying for.

Will insurance cover exome sequencing? Sometimes. It happens when a physician documents a suspected genetic condition, prior targeted testing that came back negative, and a plausible change in management. Coverage is decided case by case and prior authorization is standard. Elective personal sequencing is not covered.

What diseases can be diagnosed with whole exome sequencing? Mendelian conditions caused by point mutations, small indels, and some exon-level copy number changes in coding sequence. Examples are metabolic disorders, many neurodevelopmental and neuromuscular conditions, ciliopathies, inborn errors of immunity and hereditary cancer syndromes. It does not diagnose common complex disease, and it misses repeat expansions such as those in Huntington disease and fragile X.

Is exome sequencing worth it? For a clinician-driven rare disease workup, yes. For a technically fluent person buying their own data, a genome gives you more per dollar and no capture blind spots⁶.

Footnotes

1. Alison J Coffey, Felix Kokocinski, Maria S Calafato, et al. The GENCODE exome: sequencing the complete human exome. European Journal of Human Genetics, 2011. https://doi.org/10.1038/ejhg.2011.28 ↩

2. Murim Choi, Ute I. Scholl, Weizhen Ji, et al. Genetic diagnosis by whole exome capture and massively parallel DNA sequencing. Proceedings of the National Academy of Sciences, 2009. https://doi.org/10.1073/pnas.0910672106 ↩

3. Andrew B Singleton. Exome sequencing: a transformative technology. The Lancet Neurology, 2011. https://doi.org/10.1016/s1474-4422(11)70196-x ↩

4. Jacek Majewski, Jeremy Schwartzentruber, Emilie Lalonde, et al. What can exome sequencing do for you?. Journal of Medical Genetics, 2011. https://doi.org/10.1136/jmedgenet-2011-100223 ↩

5. Eleanor G. Seaby, Reuben J. Pengelly, Sarah Ennis. Exome sequencing explained: a practical guide to its clinical application. Briefings in Functional Genomics, 2015. https://doi.org/10.1093/bfgp/elv054 ↩

6. Aziz Belkadi, Alexandre Bolze, Yuval Itan, et al. Whole-genome sequencing is more powerful than whole-exome sequencing for detecting exome variants. Proceedings of the National Academy of Sciences, 2015. https://doi.org/10.1073/pnas.1418631112 ↩ ↩² ↩³

7. Amanda Warr, Christelle Robert, David Hume, et al. Exome Sequencing: Current and Future Perspectives. G3 Genes|Genomes|Genetics, 2015. https://doi.org/10.1534/g3.115.018564 ↩

8. Zuoheng Wang, Xiangtao Liu, Bao-Zhu Yang, et al. The Role and Challenges of Exome Sequencing in Studies of Human Diseases. Frontiers in Genetics, 2013. https://doi.org/10.3389/fgene.2013.00160 ↩