Whole Genome Sequencing Review: What You Get, What It Costs, What It Misses

Whole genome sequencing gives you a 30x short-read Illumina dataset covering roughly 3.1 Gb of reference sequence, from which a standard pipeline calls around 4–5 million SNVs and small indels against GRCh38. It costs $300–$1,500 as a consumer product and $2,000–$5,000 as a clinical test with interpretation. The value of the purchase is almost entirely determined by three things: coverage depth, whether you get the raw FASTQ and BAM/CRAM files, and whether the alignment is on GRCh38 or a pangenome-aware reference. If a vendor will not hand you the raw reads, the product is a report, not a genome, and you cannot re-analyze it when tools improve.

What 30x buys you

“30x” means the mean number of reads spanning a given base. It is a mean over a distribution that is far from uniform. GC-rich promoter regions, homopolymer tracts, and segmental duplications sit in the tail. At 30x mean, roughly 5–8% of the callable genome falls below 10x. That is about the floor where a heterozygous variant call becomes unreliable. With 8 reads and a true 50/50 allele balance, the binomial probability of seeing zero alternate reads is 0.4%. Allele-balance skew from PCR and mapping bias makes the real miss rate higher.

Practical consequences:

• Homozygous reference calls in low-coverage regions are not evidence of absence. A “negative” WGS result for a specific gene means nothing unless you check the per-base depth in that gene’s exons.
• 30x is the right target for germline SNV/indel work. Going to 50x buys you modest gains in indel sensitivity and meaningful gains for somatic or mosaic variant detection, where a variant at 5% allele fraction needs deep coverage to distinguish from sequencing error.
• Short reads (2x150bp) cannot resolve repeat expansions, large structural variants with breakpoints in repetitive sequence, or the HLA and KIR loci without specialized tools.

WGS covers coding and noncoding regions with more uniform depth than exome capture, which is why it has been displacing exomes in diagnostic settings. The clinical literature is consistent on this: genome sequencing captures the coding region more evenly than targeted capture and adds structural and noncoding variation an exome misses.¹

The files you should insist on

Ask the vendor for this list before you pay:

• FASTQ (gzipped), paired-end, with original quality scores. Roughly 100 GB for a 30x human genome. This is the only thing that lets you re-run alignment from scratch against a new reference.
• CRAM aligned to GRCh38, with the reference build and accession stated (GCA_000001405.15, analysis set with or without ALT contigs matters). CRAM is 40–60 GB versus 90–120 GB for BAM and is lossless if reference-based compression is configured correctly.
• gVCF as well as the filtered VCF, since the filtered file alone is not enough. The gVCF carries reference-block records with depth and GQ, so you can distinguish “confidently homozygous reference” from “no data here.” Without it, you are guessing.
• The pipeline manifest. It names the aligner and version (BWA-MEM2 vs DRAGEN) and the caller (GATK HaplotypeCaller, DeepVariant, DRAGEN). It also states duplicate marking and whether BQSR is on or off.

If a vendor charges a monthly fee to keep your data accessible, download everything on day one and store it yourself. A 30x genome in CRAM plus gVCF is under 70 GB, which is a few dollars a month in object storage.

Which provider

We would rank the options this way.

Nebula Genomics and Dante Labs sell 30x consumer WGS in the $300–$900 range and both deliver FASTQ, BAM, and VCF. Turnaround is the weak point, with multi-month delays reported often enough to plan around. Sequencing.com resells and adds a subscription layer over the data, which is the wrong architecture for someone who wants to own the files.

For clinical work, GeneDx, Invitae, and Baylor Genetics run CAP/CLIA labs with variant curation by board-certified geneticists and orthogonal confirmation by Sanger for reported variants. Cost is $2,000–$5,000, often billed through insurance with a physician order. You get a curated report, which is the part that carries clinical weight, and you should ask specifically whether raw data is released.

Our recommendation for a technically fluent reader: buy a research-grade 30x run where you receive FASTQ, and if anything actionable turns up, get it confirmed in a CLIA lab through a clinician. Consumer WGS variant calls are not a diagnosis and should never be treated as one. A pathogenic call in BRCA1 or a cardiomyopathy gene from a non-clinical pipeline needs orthogonal confirmation and a genetic counselor before it means anything about your health.

Reprocessing the data yourself

The pipeline we would run on a modern workstation or a single cloud instance:

bwa-mem2 mem -t 32 -R '@RG\tID:1\tSM:sample\tPL:ILLUMINA' \
  GRCh38_full_analysis_set_plus_decoy_hla.fa r1.fq.gz r2.fq.gz \
  | samtools sort -@ 8 -o sample.bam -
samtools index sample.bam

Then call with DeepVariant rather than GATK if you want fewer hand-tuned filters:

run_deepvariant --model_type=WGS \
  --ref=GRCh38_full_analysis_set_plus_decoy_hla.fa \
  --reads=sample.bam --output_vcf=sample.vcf.gz \
  --output_gvcf=sample.g.vcf.gz --num_shards=32

Use the full analysis set with decoy and HLA contigs. Aligning to a reference without decoys pushes reads that belong to unplaced sequence onto the wrong homologous locus and generates false heterozygous calls, a known and avoidable failure mode.

Annotate with Ensembl VEP, adding gnomAD v4 allele frequencies, ClinVar, and CADD:

vep -i sample.vcf.gz --cache --assembly GRCh38 --everything \
  --plugin CADD,whole_genome_SNVs.tsv.gz \
  --custom gnomad.genomes.v4.sites.vcf.bgz,gnomADg,vcf,exact,0,AF \
  -o sample.vep.vcf --vcf

Then filter hard. Start with variants at gnomAD AF < 0.001 with a ClinVar pathogenic or likely-pathogenic assertion and a review status of at least two stars. That takes 4.5 million variants down to a handful. Most of what remains will be carrier status for recessive conditions, which is informative for reproductive planning and largely silent otherwise.

For structural variants, run Manta and a read-depth caller like CNVnator and take the intersection. Short-read SV calls have high false positive rates and the consensus of two orthogonal signals is the cheapest way to cut them. For repeat expansions in genes like HTT, FMR1, and C9orf72, use ExpansionHunter with the known-locus catalog, and understand that short reads underestimate expansion length once the repeat exceeds the fragment size.

The real downsides

Interpretation is the bottleneck. Sequencing is not the limiting step. A typical genome carries thousands of rare variants of uncertain significance, and the fraction of noncoding variation anyone can interpret functionally remains small. The field’s own framing is that sequencing capacity has outrun our ability to map genotype to phenotype, and closing that gap requires functional assays and population-scale association data rather than more reads.²

Second, the reference bias problem. GRCh38 is one linear sequence, and variants in individuals whose ancestry is poorly represented in the reference and in the variant databases get called and classified worse. ClinVar and gnomAD skew European, so the same variant is more likely to be labeled “uncertain significance” for a non-European genome.

Third, incidental findings. A genome will tell you things you did not ask about, including risk alleles for adult-onset conditions with no established preventive action. Decide what you want to know before you look, and take anything medically relevant to a clinician rather than acting on it yourself.

Fourth, a genome is static. It is the same at 20 and at 70. It tells you about constraint and predisposition without describing current state. That is why sequencing is most useful read alongside data that changes: expression, protein levels, metabolic markers. Cost declines have made routine genome sequencing feasible at scale, and the useful next question is what you measure on top of it.³

Questions people also ask

Which company is the best for whole genome sequencing? For clinical answers, a CAP/CLIA lab like GeneDx or Baylor with a physician order and curated reporting. For data ownership at low cost, Nebula or Dante Labs at 30x with FASTQ delivery, accepting slower turnaround. Avoid any vendor that gates your own files behind a subscription.

What are the downsides of whole genome sequencing? Variants of uncertain significance dominate the output, and short reads miss repeat expansions and many structural variants. Coverage is uneven, so a negative result is not proof of absence. Ancestry bias in reference databases degrades interpretation for non-European genomes. Genome sequencing in clinical practice also raises consent and incidental-finding questions that are handled by genetic counselors for a reason.¹

What is the average cost of whole genome sequencing? $300–$1,500 for consumer 30x, $2,000–$5,000 for a clinical test with interpretation and Sanger confirmation. Reagent cost per genome on current high-throughput instruments is well under $200, so most of what you pay covers library prep labor, compute, storage, and interpretation.

Is 30x enough? For germline SNVs and small indels, yes. For somatic or mosaic variants below about 10% allele fraction, no. Go to 60x or higher, or use targeted deep sequencing of the region in question.

Footnotes

1. Frederik Otzen Bagger, Line Borgwardt, Andreas Sand Jespersen, et al. Whole genome sequencing in clinical practice. BMC Medical Genomics, 2024. https://doi.org/10.1186/s12920-024-01795-w ↩ ↩²

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. Sang Tae Park, Jayoung Kim. Trends in Next-Generation Sequencing and a New Era for Whole Genome Sequencing. International Neurourology Journal, 2016. https://doi.org/10.5213/inj.1632742.371 ↩