What Genetic Testing for Disease Can and Cannot Tell You

Genetic testing for disease answers three separate questions, and most confusion comes from treating them as one. First: do you carry a rare, high-penetrance variant that by itself causes or strongly predisposes to a specific condition? The genes in question include BRCA1/2, LDLR and MYBPC3. They also include HFE, CFTR and the Lynch syndrome genes. Second: do you carry common variants that change how you metabolize certain drugs? The genes here are CYP2C19, CYP2D6, DPYD, TPMT and SLCO1B1. Third: what is your aggregate common-variant burden for a polygenic trait like coronary artery disease or type 2 diabetes? The first is close to a yes/no answer with real clinical weight for a small number of people. The second is a lookup table with strong evidence behind it. The third is a population-level probability distribution that shifts your prior modestly and is heavily dependent on the ancestry of the training cohort. A test that is excellent at one of these can be useless at the others.

The three questions, and what each is worth

Monogenic disease variants. Roughly 2–3% of unselected adults carry a pathogenic or likely pathogenic variant in one of the ACMG secondary-findings genes. The list is now 81 genes in v3.2. It covers hereditary cancer syndromes, cardiomyopathies and arrhythmias, and familial hypercholesterolemia. It also covers malignant hyperthermia susceptibility and a handful of others. These are the findings that change what a clinician does. If you get one, you need a genetic counselor and a confirmatory clinical-grade test. A spreadsheet is not enough. Penetrance is the thing people get wrong: a pathogenic BRCA1 variant found in a woman with three affected first-degree relatives carries much higher lifetime risk than the same variant found incidentally in someone with no family history, because ascertainment bias inflates the published penetrance estimates.

Pharmacogenomics. This is the highest-yield, lowest-drama category. Star-allele calls for a couple dozen genes with CPIC guidelines behind them. Nearly everyone carries at least one actionable variant. Note that we are describing measurement: the guidelines themselves are clinician-facing, and dosing decisions belong with your prescriber.

Polygenic scores. A CAD polygenic score built from ~1.7M variants will separate the top decile from the median by a meaningful hazard ratio in European-ancestry cohorts and by considerably less in others. It is a modifier on top of age, sex, lipids, blood pressure, and family history. The score does not replace them. Treat a PRS as you would a noisy feature in a model where the other features are already strong.

Array versus exome versus whole genome

Consumer arrays (23andMe, AncestryDNA) genotype 600k–700k pre-selected sites. They are good at what they measure and blind to everything else. Two failure modes matter. First, arrays only interrogate variants on the chip, so a “negative” BRCA result from an array that tests three Ashkenazi founder variants says almost nothing about the thousands of other pathogenic BRCA alleles. Second, rare-variant calls on arrays have poor positive predictive value: a large validation study found that a substantial fraction of rare pathogenic-looking calls in consumer raw data were false positives on Sanger confirmation. Never act on an array-derived rare variant without orthogonal confirmation.

Exome sequencing covers ~1–2% of the genome, the protein-coding part, where the majority of known monogenic disease variants sit. It misses deep intronic splice variants, most structural variation, and has uneven capture at GC-rich first exons.

Whole-genome sequencing at 30x is what we would choose. It gives uniform coverage, calls SNVs, indels, CNVs, and structural variants from a single library, and captures repeat expansions and mitochondrial variants that capture-based methods handle badly. The diagnostic argument for WGS over exome in clinical settings is now well documented, both in yield and in the ability to reanalyze the same data as knowledge improves ¹. The broader point is that the marginal cost of sequencing everything has fallen far enough that targeted panels are increasingly an interpretive choice rather than an economic one ². The policy and ethics literature has been arguing about how to handle the incidental findings that come with that breadth since 2013, and that argument is not settled ³.

What you get and what to run on it

A 30x WGS deliverable should include:

• FASTQ (raw reads, ~100 GB gzipped for 30x paired-end 150bp)
• CRAM or BAM aligned to GRCh38 (CRAM is ~60% the size, use it)
• gVCF and a filtered VCF
• The reference build and aligner version in writing

Check quality before anything else:

samtools stats sample.cram | grep ^SN
mosdepth --by 500 --fast-mode sample sample.cram
bcftools stats sample.vcf.gz

These are the numbers you want.

• mean depth ≥30x
• ≥95% of the callable genome at ≥20x
• duplicate rate under ~15%
• Ti/Tv around 2.0–2.1 genome-wide (higher, ~3.0, restricted to coding)
• 4.5–5M total variants for a typical genome

A het/hom ratio far from expectation for your ancestry usually means contamination. Run verifyBamID2 if you suspect it.

For annotation, we use Ensembl VEP with the --everything flag plus ClinVar and gnomAD v4:

vep -i sample.vcf.gz --cache --assembly GRCh38 --everything \
  --custom clinvar.vcf.gz,ClinVar,vcf,exact,0,CLNSIG,CLNREVSTAT \
  --custom gnomad.genomes.v4.1.sites.vcf.bgz,gnomADg,vcf,exact,0,AF \
  --vcf -o annotated.vcf.gz --compress_output bgzip

Then filter hard. The single most useful filter: population frequency. A variant causing a dominant disease with meaningful penetrance cannot be at 1% allele frequency in gnomAD. Keep gnomAD AF < 0.001 for dominant conditions, < 0.01 for recessive. Then require ClinVar CLNSIG of Pathogenic or Likely_pathogenic with CLNREVSTAT of at least two submitters with no conflicts, restricted to the ACMG SF v3.2 gene list. That pipeline typically returns zero to three variants. Everything that survives goes to a clinician for confirmation on a CLIA-validated assay.

Do not act on variants of uncertain significance. There are hundreds in every genome and reclassification data shows most VUS eventually move to benign. Do not act on CLNSIG values with conflicting interpretations. Do not skip the confirmation step because the variant “looks obviously damaging.”

For pharmacogenomics, run PharmCAT (pharmcat_pipeline sample.vcf.gz) which handles star-allele calling and emits CPIC-guideline annotations. CYP2D6 needs structural variant awareness, and PharmCAT can call StellarPGx or Aldy for it. Take the report to whoever writes your prescriptions.

For polygenic scores, pull weight files from the PGS Catalog and score with PLINK 2:

plink2 --vcf sample.vcf.gz --score PGS000018.txt 1 4 6 header \
  list-variants cols=+scoresums

The number is meaningless without a reference distribution. Score a matched-ancestry cohort (1000 Genomes works) and report your percentile within it. If your ancestry is not represented in the score’s development cohort, say so in the output rather than reporting a percentile that does not apply.

Questions people also ask

What diseases can you get genetic testing for? Thousands. The high-confidence ones are single-gene conditions with established gene-disease validity.

• hereditary breast and ovarian cancer
• Lynch syndrome
• familial hypercholesterolemia
• hypertrophic and dilated cardiomyopathy
• long QT syndrome
• hereditary hemochromatosis
• cystic fibrosis
• sickle cell disease
• Huntington disease
• the ~700 genes in carrier screening panels

Testing for polygenic conditions like type 2 diabetes exists but gives a probability shift. It does not give an answer.

Which disease would genetic testing be most likely to identify? In an unselected adult, hereditary hemochromatosis (HFE C282Y homozygosity, ~1 in 200 in Northern European ancestry) and familial hypercholesterolemia (~1 in 250) are the most common findings that matter clinically.

What is the most common genetic disorder? By carrier frequency, hemochromatosis and FH lead among adults of European ancestry. Globally, the hemoglobinopathies (sickle cell disease and the thalassemias) affect the most people.

Can a consumer DNA test check for diseases? Partially. Array-based tests cover a fixed handful of variants per gene and produce false positives on rare calls. A negative result from one is weak evidence. If you want a genome-wide answer, sequence the genome.

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. 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. 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 ↩