What to Look For in a Whole Genome Sequencing Service

Oak
wgs genomics sequencing bioinformatics vcf
A small feathered reptile specimen on a black plinth, its feathers made of many stacked translucent layers lit in teal and amber.

A whole genome sequencing service takes a blood or saliva sample and returns the sequence of roughly 3.1 billion base pairs, usually as a FASTQ or BAM/CRAM file plus a variant call file. The thing that distinguishes a good service from a bad one is not the sequencing chemistry, which is largely commoditized, but three questions: what coverage depth you get, whether you receive the raw reads, and whether the variant calls come with the alignment and the quality metrics needed to check them. Consumer prices for 30x short-read WGS with raw data delivery currently sit between roughly $300 and $1,000. Clinical WGS with a physician-signed interpretive report runs several thousand dollars, because you are paying for the interpretation and the regulatory apparatus, not the reads.

If you are technically fluent and want to work with the data yourself, buy the reads.

What you should receive, file by file

A complete delivery looks like this:

  • sample_R1.fastq.gz and sample_R2.fastq.gz, paired-end reads, typically 2x150 bp. For a 30x human genome this is about 90-110 GB compressed.
  • sample.bam or sample.cram, aligned to a named reference. CRAM against GRCh38 is roughly 15-20 GB; the equivalent BAM is 60-80 GB. CRAM is reference-dependent, so make sure you know exactly which FASTA was used, including whether it carries alt contigs and the HLA decoys.
  • sample.vcf.gz plus .tbi index, small variants (SNVs and indels).
  • Ideally a structural variant VCF from a caller like Manta or Delly, and a copy-number file.
  • A QC report: mean coverage, percent of the genome at ≥10x and ≥20x, duplicate rate, insert size distribution, Q30 percentage, contamination estimate.

If a provider will not tell you the reference build before you order, do not order. GRCh38 is the right answer in 2026. Some services still ship GRCh37/hg19 because their downstream annotation pipeline was built around it, which means you will spend an afternoon with CrossMap or bcftools +liftover before you can use most modern annotation sources. T2T-CHM13 is better as a reference in absolute terms, especially in centromeric and segmental-duplication regions, but the annotation ecosystem around it is still thinner, so we run GRCh38 as primary and CHM13 as a secondary alignment when a region of interest falls in a known GRCh38 gap.

Coverage: 30x is the floor, and it is an average

“30x” means the mean number of reads covering a base. The distribution matters more than the mean. GC-rich promoters and first exons routinely fall to 10x or below in PCR-based libraries. Ask for PCR-free library prep. It costs the provider slightly more and it flattens GC bias substantially, which shows up directly in your callable fraction.

Check the delivered data yourself rather than trusting the report:

samtools coverage -H sample.cram | awk '{cov+=$6*$3} END {print cov}'
mosdepth --by 1000 --fast-mode sample sample.cram

mosdepth will give you per-base and windowed depth in a few minutes. What you want to see: ≥95% of the callable genome at ≥20x, duplicate rate under about 10%, and a contamination estimate (from VerifyBamID2) under 0.01. A contamination estimate above 0.02 means heterozygous calls are unreliable and you should ask for a re-run.

For a single germline baseline, 30x PCR-free short read is the sensible default. Going to 60x buys you modest gains in indel sensitivity and not much else for germline SNVs. If you care about somatic variants, structural variation, or repeat expansions, depth is the wrong lever; you want long reads.

Short reads, long reads, and what each one misses

Short-read Illumina sequencing at 30x calls SNVs in the mappable genome at sensitivity and specificity above 99%. It is weak in three places: repeat expansions (Huntington’s CAG tract, FMR1 CGG, C9orf72 GGGGCC), large structural rearrangements, and phasing. Read length sets a hard ceiling there, and no amount of depth fixes it. The technology improvements that made 30x genomes cheap were improvements in throughput and read count, not read length 1.

Oxford Nanopore and PacBio HiFi solve those problems. PacBio HiFi at 15-20x gives you accurate SNVs plus structural variants plus native phasing into haplotype blocks that often span whole genes. Nanopore gives you the same structural resolution plus direct 5mC methylation calls from the raw signal, no bisulfite conversion. The tradeoff is price (still several times short-read cost per sample) and a smaller set of population reference data to compare against.

Our view: short-read 30x PCR-free is the right first genome for almost everyone, because the annotation and population-frequency ecosystem is built around it. Add long reads later if a specific question demands it.

Interpretation is where the difficulty lives

Sequencing is solved. Turning a 4-5 million-variant VCF into something meaningful is not.

A typical individual genome carries roughly 4-5 million variants relative to the reference, of which perhaps 10,000-12,000 are protein-altering and a few hundred are rare and predicted damaging. Filtering that to anything actionable requires gnomAD frequencies, ClinVar assertions, and constraint metrics, and even then the residual set is mostly variants of uncertain significance. In the original Stanford study applying clinical interpretation to whole genomes, manual curation of each genome took roughly 100 hours of expert time, and reviewers disagreed on which variants warranted reporting 2. That gap has narrowed with better databases, but it has not closed.

The practical consequences are worth stating plainly. Reported findings from WGS in unselected adults are dominated by carrier status and pharmacogenomic variants, not by high-penetrance disease alleles. Penetrance estimates for many ClinVar “pathogenic” variants come from ascertained families and are substantially lower in unselected populations. And reanalysis matters: a variant classified as uncertain in 2026 may be reclassified in 2029, which is an argument for holding the raw data rather than a static PDF 3.

If a variant in your data has clinical implications, that is a conversation with a genetic counselor or physician, working from a CLIA-validated confirmatory test. Research-grade WGS calls are not a diagnosis, and a single-sample VCF entry at 12x depth with an allele balance of 0.25 is often an artifact.

A first pass on your own data

Assuming you have sample.vcf.gz against GRCh38:

# annotate
bcftools norm -m-any -f GRCh38.fa sample.vcf.gz -Oz -o norm.vcf.gz
bcftools index norm.vcf.gz
vep -i norm.vcf.gz --cache --assembly GRCh38 \
    --everything --af_gnomadg --canonical --vcf \
    -o annotated.vcf --fork 8

# rare, protein-altering, high-quality
bcftools filter -i 'QUAL>50 && FMT/DP>15' annotated.vcf \
  | filter_vep --filter "gnomADg_AF < 0.001 and IMPACT is HIGH"

That last filter typically leaves a few hundred variants. Cross-reference against ClinVar (clinvar.vcf.gz, updated weekly) and the ACMG secondary-findings gene list, which is a defined set of genes where incidental findings are considered reportable. Separately, run the PharmGKB/CPIC-curated star alleles through PyPGx or Aldy, which handle the copy-number complexity of CYP2D6 properly. A naive VCF lookup will get CYP2D6 wrong.

Questions people also ask

How much does a WGS test cost? Consumer 30x short-read WGS with raw data delivery runs roughly $300-$1,000 depending on turnaround and whether the library is PCR-free. Clinical WGS with a signed interpretive report from a CLIA/CAP lab typically runs $3,000-$10,000, and the difference is interpretation and regulatory overhead rather than better sequencing.

Is the sequencing company legitimate? Check three things: the CLIA number and CAP accreditation if you want clinically usable results, whether they name the sequencing platform and reference build up front, and whether raw FASTQ and BAM/CRAM are included or an upsell. A provider that will not name the instrument or the reference is reselling capacity and hiding the details.

What are the benefits of whole genome sequencing over an exome or a SNP array? An array genotypes 500,000-900,000 pre-selected positions and imputes the rest, so it cannot see rare variants it was not designed for. An exome covers about 1-2% of the genome and misses deep intronic and regulatory variants. WGS gives uniform coverage across coding and non-coding regions, better structural-variant sensitivity, and mitochondrial DNA, in a single assay 3.

What is the purpose of WGS? For an individual, it is a one-time permanent baseline: a file that does not expire and that gets more informative as annotation databases improve. The sequence does not change, so the same FASTQ reanalyzed in five years against better references and larger population cohorts will tell you more than it does today.

Oak builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.

Footnotes

  1. 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 ↩

  2. Frederick E. Dewey, Megan E. Grove, Cuiping Pan, et al. Clinical Interpretation and Implications of Whole-Genome Sequencing. JAMA, 2014. https://doi.org/10.1001/jama.2014.1717 ↩

  3. 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 ↩ ↩2