Is 100X whole genome sequencing worth it?

For a standard germline question (what inherited variants do I carry), 30x whole genome sequencing answers it and 100x does not answer it much better. Depth beyond roughly 30x improves SNV and indel calling only marginally, because a heterozygous site at 30x already yields around 15 alternate reads and callers saturate well before that. Deep coverage earns its cost when the variant you care about is present in a minority of your cells: post-zygotic mosaicism, mitochondrial heteroplasmy, clonal hematopoiesis, or tumor DNA circulating in plasma. If that is your question, 100x is a floor rather than a luxury, and 200x or more is often the right number.

What depth means, and why the mean is the wrong summary

“100X” describes the mean number of reads covering each base after alignment and duplicate marking. The mean hides the distribution, which is what determines whether a specific locus is callable. Coverage in practice is overdispersed and structured by GC content, mappability, and library chemistry. The distribution is not Poisson.

The consequence is that a nominal 30x genome typically leaves a few percent of bases below any usable threshold. GC-rich first exons, segmental duplications, and homopolymer tracts drop out first. Raising the mean to 100x lifts many of those regions above threshold, but it does not fix regions that fail because reads cannot be uniquely placed. For mappability-limited loci, longer reads or linked-read and long-read data are the fix. More short reads do not help. Experimentally phased personal genomes make the same point: haplotype resolution comes from library design rather than depth alone.¹

Ask any vendor whether the advertised number is raw, mapped, or deduplicated. A “100x” figure quoted on raw gigabases can become 80x or less after duplicate removal and soft-clipping. Around 100 Gb of aligned 2x150 bp data gives roughly 30x on a 3.1 Gb reference.

What 30x already gives you

At 30x, single-nucleotide variant calling is close to its ceiling. Modern callers on PCR-free libraries reach SNV sensitivity and precision above 99% against benchmark truth sets at this depth. Platform comparison studies using Genome in a Bottle reference samples report SNV F-scores in that range for standard 30x to 50x WGS.² Indels remain harder, particularly in homopolymers, and that difficulty comes from alignment ambiguity rather than a lack of reads.

Clinical laboratories validate their pipelines at similar depths. Analytical validation work for clinical exome and genome sequencing sets minimum per-base depth thresholds in the range of 10x to 20x for a callable site, with overall genome coverage targeted well above that.³ Diagnostic yield studies in severe pediatric disease used genomes in this range and found the limiting factor to be interpretation. Read count was not the limit.⁴

The practical implication is straightforward. If you want your polygenic scores, pharmacogenomic star alleles, carrier status, and rare coding variants, 30x PCR-free WGS with a good aligner and caller is the efficient choice. Spending the same budget on RNA sequencing, proteomics, or repeated blood biomarkers usually tells you more than tripling germline depth.

What 100x buys

Deep coverage changes what fraction of variants you can detect at low allele fraction. The arithmetic is binomial. Consider a mosaic variant present in 5% of cells, so about 5% of reads carry the alternate allele. At 30x, the probability of seeing at least three alternate reads is about 19%. At 100x, it is about 88%. Three alternate reads is a common minimum before a caller will emit anything above noise.

That single calculation explains most of the legitimate uses of 100x and deeper:

• Somatic and mosaic variant detection in blood or tissue. Statistical frameworks for somatic calling model the per-site error distribution across a panel of normals specifically because low-VAF calls are dominated by sequencing artifacts.⁵
• Cell-free DNA in plasma, where tumour fraction can be under 1%. Deep WGS of cfDNA can detect copy number and mutational signal at low tumour content, but sensitivity falls steeply as fraction drops.⁶
• Mitochondrial heteroplasmy. The mitochondrial genome is already sequenced at 100x to 1000x in a nuclear 30x run because of copy number per cell, so this often needs no extra depth. Check the actual chrM depth before assuming.
• Structural variant and copy number calling, where breakpoint-spanning reads are scarce and extra depth improves both split-read and discordant-pair evidence.

Note the asymmetry. Cell-free DNA applications and tumour sequencing routinely use depths in the hundreds or thousands, so 100x is a compromise rather than a solution there.⁷ Choosing depth means choosing a target allele fraction first, then solving for the reads you need.

Checking the data you were sent

Whatever depth you buy, verify it yourself. Require FASTQ or CRAM as well as a VCF. A VCF alone is not enough. A VCF is a lossy summary produced by someone else’s pipeline choices, and you cannot recall variants from it.

Start with coverage distribution rather than the mean:

mosdepth --by 1000 --fast-mode --threads 4 sample sample.cram
# then look at sample.mosdepth.global.dist.txt:
# fraction of genome at >=10x, >=20x, >=30x

For a genuine 100x PCR-free library you should see well over 95% of the non-N genome at 30x or more. Then check library quality:

samtools stats sample.cram > stats.txt
grep ^SN stats.txt | cut -f2-
# insert size average should be ~350-450 bp
# duplicate rate should be under ~10% for PCR-free

A high duplicate rate means the vendor amplified a low-input library and your effective depth is lower than advertised. High duplication also creates correlated errors that look like low-VAF variants, which is exactly the signal you paid extra to see.

For calling, our default is BWA-MEM2 to GRCh38 with alt-aware handling, then DeepVariant with --model_type=WGS for germline SNVs and indels. For anything mosaic, use a somatic caller with a panel of normals. Benchmark datasets built across centers and platforms exist precisely so you can measure your own false positive rate before trusting a low-VAF call.⁸ For mitochondrial heteroplasmy, GATK Mutect2 in --mitochondria-mode followed by haplocheck for contamination is the path we use.

One caveat applies to every low-fraction call. A variant seen at 3% VAF in one sample and never confirmed is a hypothesis. It is not a finding. Confirmation requires an orthogonal assay in a clinical laboratory, and interpretation of anything medically relevant requires a clinician or genetic counselor.

How we would choose

Buy 30x PCR-free WGS with raw data delivery for the germline question. Buy 100x or deeper only when you have named the low-fraction signal you are chasing and computed the depth it requires. A mixed design is often better value than uniform deep WGS: deep targeted or exome sequencing for regions where coding variants concentrate, combined with lower-depth genome-wide data for structural and copy number context.⁹ Depth is one axis of a budget, and it competes with the other molecular layers that change over time.

Questions people also ask

How much does it cost to get a full genome sequencing test?

Consumer 30x whole genome sequencing is commonly priced between a few hundred and roughly a thousand dollars, with 100x offerings typically two to four times that. Clinical-grade sequencing ordered through a physician, including interpretation and confirmatory testing, costs substantially more.

Is whole genome testing worth it?

It is worth it if you will use the raw data. Most healthy adults will find a handful of carrier variants, pharmacogenomic alleles, and polygenic scores of modest effect. The diagnostic yield is much higher in symptomatic disease, where genome sequencing resolves a meaningful fraction of previously undiagnosed neurodevelopmental cases.⁴

Does insurance cover whole genome testing?

Coverage generally requires a clinical indication and a physician order, such as a suspected genetic disorder or an undiagnosed condition. Elective sequencing for a healthy adult is typically paid out of pocket.

Which whole genome sequencing test is the best?

Choose the one that delivers PCR-free library preparation and 2x150 bp reads. It should also give you real 30x or better deduplicated coverage and unrestricted FASTQ or CRAM download. Judge providers on deliverables and quality metrics rather than on the reports they bundle, since you can regenerate any report from the raw data.

Footnotes

1. Qing Mao, Serban Ciotlos, Rebecca Yu Zhang, et al. The whole genome sequences and experimentally phased haplotypes of over 100 personal genomes. Gigascience, 2016. https://doi.org/10.1186/s13742-016-0148-z ↩

2. Chaoyang Li, Xue Fan, Xin Guo, et al. Accuracy benchmark of the GeneMind GenoLab M sequencing platform for WGS and WES analysis. BMC Genomics, 2022. https://doi.org/10.1186/s12864-022-08775-3 ↩

3. Michael D Linderman, Tracy Brandt, Lisa Edelmann, et al. Analytical validation of whole exome and whole genome sequencing for clinical applications. BMC Medical Genomics, 2014. https://doi.org/10.1186/1755-8794-7-20 ↩

4. Sarah E. Soden, Carol J. Saunders, Laurel K. Willig, et al. Effectiveness of exome and genome sequencing guided by acuity of illness for diagnosis of neurodevelopmental disorders. Science Translational Medicine, 2014. https://doi.org/10.1126/scitranslmed.3010076 ↩ ↩²

5. Yuichi Shiraishi, Yusuke Sato, Kenichi Chiba, et al. An empirical Bayesian framework for somatic mutation detection from cancer genome sequencing data. Nucleic Acids Research, 2013. https://doi.org/10.1093/nar/gkt126 ↩

6. Devika Ganesamoorthy, Alan James Robertson, Wenhan Chen, et al. Whole genome deep sequencing analysis of cell-free DNA in samples with low tumour content. BMC Cancer, 2022. https://doi.org/10.1186/s12885-021-09160-1 ↩

7. Rossa W. K. Chiu, K. C. Allen Chan, Yuan Gao, et al. Noninvasive prenatal diagnosis of fetal chromosomal aneuploidy by massively parallel genomic sequencing of DNA in maternal plasma. Proceedings of the National Academy of Sciences, 2008. https://doi.org/10.1073/pnas.0810641105 ↩

8. Yongmei Zhao, Li Tai Fang, Tsai-wei Shen, et al. Whole genome and exome sequencing reference datasets from a multi-center and cross-platform benchmark study. Scientific Data, 2021. https://doi.org/10.1038/s41597-021-01077-5 ↩

9. Claude Bhérer, Robert Eveleigh, Katerina Trajanoska, et al. A cost-effective sequencing method for genetic studies combining high-depth whole exome and low-depth whole genome. npj Genomic Medicine, 2024. https://doi.org/10.1038/s41525-024-00390-3 ↩