Genome Screening: What a Whole Genome Tells You

Genome screening means sequencing a person who has no specific clinical question and looking across their DNA for variants with known consequences. In a healthy adult, the yield is small and well-characterized. Roughly 1 to 3 percent of people carry a pathogenic or likely-pathogenic variant in the ~81 genes on the ACMG secondary findings list (v3.2). Most people carry two to four recessive carrier variants, and essentially everyone carries a handful of pharmacogenomic star alleles that change how a drug is metabolized. Everything else in your VCF is either common, benign, or uninterpretable. The interesting part for a technical reader is the data itself: 30x PCR-free WGS gives you a permanent, re-analyzable artifact that gets more informative every year as ClinVar and gnomAD grow, without resequencing.

That said, screening a healthy person is genuinely different from diagnosing a sick one, and the professional literature has been consistent that the two should not be conflated. European guidance recommends targeted analysis of the sequence where a clinical question exists, rather than opportunistic interrogation of the whole file, precisely because unsolicited findings in asymptomatic people have poorly characterized penetrance¹. A Mayo Clinic review of WGS in healthy people makes the sharper point: a variant annotated “pathogenic” in a patient ascertained through a family with disease does not carry the same predictive weight when found in someone with no phenotype and no family history².

What you should get back, as files

Insist on these, in this order of importance:

• FASTQ or unaligned BAM. The raw reads. If you have these you can rerun any aligner and any caller for the rest of your life. ~90–110 GB gzipped for a 30x human genome.
• CRAM aligned to GRCh38 (with ALT contigs handled, not stripped), plus the .crai. ~50–60 GB. Ask which reference FASTA was used and get its MD5. GRCh38_full_analysis_set_plus_decoy_hla.fa is the sane default.
• gVCF and a filtered VCF. A single genome yields roughly 4.5–5.0 million SNVs and 500,000–800,000 indels against GRCh38. The gVCF matters because it distinguishes “reference” from “no coverage,” which a plain VCF does not.
• SV and CNV calls (Manta, or a graph caller) and repeat expansion calls (ExpansionHunter). QC: mosdepth summary, verifybamid2 FREEMIX and insert size distribution.

If a provider will only hand you a PDF report and a spreadsheet of 40 variants, you have bought an interpretation. You have not bought a genome.

The pipeline we would run

Alignment with bwa-mem2 mem -K 100000000 -Y (the -K makes output deterministic regardless of thread count), or DRAGEN if you have access. Small variants with DeepVariant (--model_type=WGS), which we prefer over GATK HaplotypeCaller for single-sample work because its indel precision in homopolymers is better and it needs no BQSR step. Then annotate:

vep --offline --cache --assembly GRCh38 --everything \
    --fasta GRCh38.fa --vcf --fork 8 \
    --plugin dbNSFP,dbNSFP4.7a.gz,ALL \
    --plugin SpliceAI,snv=spliceai_snv.hg38.vcf.gz,indel=... \
    --custom gnomad.joint.v4.1.sites.vcf.gz,gnomADj,vcf,exact,0,AF_joint \
    --custom clinvar.vcf.gz,ClinVar,vcf,exact,0,CLNSIG,CLNREVSTAT

Then filter hard. The first pass we run starts with ClinVar CLNSIG of Pathogenic or Likely_pathogenic, with CLNREVSTAT at two stars or better (criteria_provided,_multiple_submitters,_no_conflicts or a practice guideline). It then requires gnomAD v4 joint allele frequency below 0.001 for dominant conditions. Last, it is restricted to a gene list you chose before you looked. That last constraint is the one people skip and then regret. Pick your list first: ACMG SF v3.2 for actionable secondary findings, ACMG carrier panels, or a specific set tied to your family history.

That filter typically leaves single digits of variants, and most will still be wrong. Two-star ClinVar entries get reclassified. Check the current ClinVar record and the submitter-level evidence. The aggregate label alone is not enough.

Separately, run PharmCAT on the VCF for pharmacogenomic diplotypes. Note that PharmCAT will not call CYP2D6 from short reads on its own because CYP2D6 and the CYP2D7 pseudogene are ~94 percent identical with recurrent hybrid alleles. Use Cyrius or StellarPGx on the CRAM for that gene and feed the result back in. The same pseudogene problem hits PMS2 (exons 11–15 vs PMS2CL), SMN1/SMN2, CYP21A2, GBA/GBAP1, and HBA1/HBA2. If a report is silent on those genes, that is a coverage artifact. It is not a clean result.

What genome screening misses

Short-read 30x WGS is blind or unreliable for:

• Repeat expansions beyond read length (HTT, FMR1, C9orf72 and RFC1). ExpansionHunter gives a genotype but long expansions saturate.
• Methylation and imprinting. Requires bisulfite sequencing or nanopore.
• Somatic mosaicism below ~5–10 percent VAF in blood, and any tissue-restricted mosaicism.
• Balanced translocations and inversions with breakpoints in repeats.
• The first exons of GC-rich genes, where coverage can drop below 10x even at 30x mean.

Check your own coverage before believing a negative: mosdepth --by targets.bed sample cram and look at the fraction of your gene list at ≥20x. Below ~95 percent of coding bases at 20x, treat “no variant found” as “not assessed.”

The risks, stated concretely

The dominant risk is a false or overstated positive. A scary true one is not the main danger. Variants of uncertain significance vastly outnumber pathogenic calls, penetrance estimates derived from affected families overstate risk in unselected populations, and a chain of confirmatory imaging, specialist visits, and anxiety can follow a variant that would never have caused disease². The second risk is false reassurance: a clean genome does not lower your risk from common polygenic disease, environment, or anything in the list above.

Policy bodies have pushed back specifically on genomic screening in contexts where consent is thin. The ESHG position on newborn screening argues that programs should stay targeted at conditions with proven, actionable interventions rather than expanding to whole genomes, because a newborn cannot consent to learning about adult-onset risk³. The same logic applies to you, voluntarily, with better information: decide in advance which categories you want returned.

Legally, in the US, GINA covers health insurance and employment. It does not cover life, disability, or long-term care insurance. If you plan to buy those policies, sequence with that timing in mind.

Anything that reaches the threshold of a pathogenic finding in a cancer-predisposition, cardiomyopathy, or arrhythmia gene needs a genetic counselor and a clinician, plus orthogonal confirmation in a CLIA laboratory. Research-grade calls are not a clinical result, and we would not act on one without that confirmation. Pharmacogenomic results should go to the prescriber. They should not go into a self-directed change.

Questions people also ask

Is genome testing covered by insurance? In the US, usually only with a clinical indication and an ordering physician, billed under CPT 81425 (genome sequencing, proband) or as an exome, 81415. Screening an asymptomatic adult is generally considered not medically necessary and is paid out of pocket. Diagnostic sequencing in a child with an unexplained neurodevelopmental or multi-system presentation is far more often approved.

How much does a full genome test cost? Reagent and instrument cost for 30x WGS is now low enough that consumer offerings sit in the few-hundred to low-thousands range, while a clinical diagnostic genome with interpretation, confirmation, and counseling is billed in the thousands to low tens of thousands. The sequencing is the cheap part. Interpretation and clinical follow-up are the cost.

What is genomic screening? Testing someone without a specific clinical suspicion to find variants with known consequences, as opposed to diagnostic testing that starts from a phenotype. The distinction changes the math: the prior probability is far lower in screening, so the positive predictive value of any given “pathogenic” call is lower too¹.

What are the risks of genomic testing? Uncertain and reclassified variants, overestimated penetrance from family-ascertained data, downstream procedures triggered by findings that would never have become disease, false reassurance from an incomplete assay, and non-health-insurance discrimination outside GINA’s scope².

Footnotes

1. 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 ↩ ↩²

2. 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 ↩ ↩² ↩³

3. Heidi Howard, Bartha Maria Knoppers, Martina C. Cornel, et al. Whole-genome sequencing in newborn screening? A statement on the continued importance of targeted approaches in newborn screening programmes. European Journal of Human Genetics, 2015. https://doi.org/10.1038/ejhg.2014.289 ↩