How to Interpret a DNA Test Result

Oak
genomics vcf variant-interpretation wgs annotation
A tall glass laboratory column filters millions of tiny blue glowing beads through fine sieves until four amber beads remain.

Interpreting a DNA test means answering four questions about each variant in the file: where it sits on the reference genome, what genotype you carry there, how common that genotype is in people like you, and what evidence exists that it changes a protein or a phenotype. Everything else is bookkeeping. The hard part is that a 30x whole genome yields roughly 4–5 million variants against GRCh38, of which perhaps 100 are rare and protein-altering in a gene with any published disease association, and typically zero to a handful are worth a clinician’s time. Interpretation is mostly a filtering problem, and the filters are where people go wrong.

What is in the file

A clinical or research sequencing result arrives as a VCF (usually bgzip’d, .vcf.gz, with a .tbi index), plus a CRAM or BAM of aligned reads. The VCF has eight fixed columns and then per-sample genotype columns:

#CHROM POS      ID          REF ALT QUAL   FILTER INFO        FORMAT      SAMPLE
chr13  32340301 rs80359550  CA  C   932.4  PASS   AC=1;AN=2   GT:AD:DP:GQ 0/1:31,28:59:99

GT is the genotype: 0/0 homozygous reference, 0/1 heterozygous, 1/1 homozygous alternate, ./. no call. AD is allelic depth (reference reads, alt reads), DP total depth, GQ a Phred-scaled genotype quality (99 is the ceiling in GATK output, 20 means a 1% chance the call is wrong). A het call with AD 31,28 looks clean. A het call with AD 40,4 at DP 44 is usually an artifact, a mismapped read pile, or somatic mosaicism, and you should open it in IGV before believing it.

Two things break more downstream analyses than anything else. First, reference build: a variant at chr17:43,094,464 in GRCh38 is at chr17:41,246,481 in GRCh37, and mixing them silently produces wrong gene assignments. Check the VCF header for ##reference and ##contig lines. Second, representation: the same indel can be written several ways. Normalize before you compare against anything.

bcftools norm -m -any -f GRCh38.fa -c x -Oz -o norm.vcf.gz raw.vcf.gz
bcftools index -t norm.vcf.gz

The filter chain we would run

Order matters. Quality first, then frequency, then consequence, then evidence.

bcftools view -f PASS -i 'FMT/DP>=10 & FMT/GQ>=20' norm.vcf.gz -Oz -o pass.vcf.gz

vep -i pass.vcf.gz -o annot.vcf --vcf --cache --assembly GRCh38 \
    --mane_select --pick_allele_gene --canonical --symbol --hgvs --numbers \
    --custom gnomad.genomes.v4.1.sites.vcf.gz,gnomADg,vcf,exact,0,AF_grpmax \
    --plugin dbNSFP,dbNSFP4.5a.gz,REVEL_score,CADD_phred \
    --plugin SpliceAI,snv=spliceai_snv.hg38.vcf.gz,indel=spliceai_indel.hg38.vcf.gz

Then cut on population frequency. For a dominant, high-penetrance hypothesis, anything above 0.1% group-max allele frequency in gnomAD is almost certainly not the cause. For recessive carrier status, 1% is a reasonable ceiling. Frequency filtering is only as good as the reference panel’s ancestry match: deep sequencing of 1,070 Japanese individuals found large numbers of rare variants absent from panels built on European cohorts, which is exactly the situation where a benign local polymorphism gets flagged as a rare candidate.1 If your ancestry is underrepresented in gnomAD, expect a longer candidate list and more false leads.

After frequency, keep loss-of-function (stop-gained, frameshift, canonical splice ±1/2), missense with REVEL above ~0.7, and anything with a SpliceAI delta score above 0.2 regardless of consequence class. Cross-reference ClinVar, but read the review status, not just the classification. A “Pathogenic” call with one submitter and no assertion criteria (zero stars) is a claim, not evidence. Two stars and up (multiple submitters, no conflicts) is worth acting on by asking a genetic counselor.

Reading one variant end to end

Take a line like NM_000059.4:c.5946del p.(Ser1982ArgfsTer22), genotype 0/1.

  • NM_000059.4 is the RefSeq transcript, version 4. Coordinates are transcript-relative, so the transcript identity is part of the answer. Use the MANE Select transcript unless the lab specified another.
  • c.5946del means a single-base deletion at coding position 5946, counting from the A of the initiator ATG.
  • p.(Ser1982ArgfsTer22) is the predicted protein effect: serine 1982 becomes arginine, and a frameshift terminates 22 residues later. The parentheses mean predicted, not observed at the protein level.
  • 0/1 means one copy. For an autosomal dominant cancer predisposition gene, one copy is the finding. For a recessive condition, one copy is carrier status and nothing more unless a second variant is in trans on the other allele, which you cannot determine from a single short-read VCF without phasing or parental samples.

Penetrance is the step most people skip. Carrying a variant is a statement about your genome. Whether and when it produces a phenotype depends on modifiers, sex, and environment. Some inherited conditions are conditionally expressed: an estrogen-dependent form of hereditary angioedema presents only in women and can be triggered by exogenous estrogen, with the same variant silent in male carriers.2 Reading a variant as a fixed prediction of your future is a misreading.

What a negative result does not rule out

“No pathogenic variants identified” is a statement about what the assay can see. Short-read sequencing at 30x calls SNVs and small indels well across most of the genome and calls almost nothing well in a few specific places.

  • Repeat expansions. Triplet and other tandem repeat disorders (HTT, FMR1, C9orf72, DMPK, ATXN genes) involve alleles longer than a 150 bp read, and standard callers miss them entirely.3 ExpansionHunter with a locus catalog is the minimum, and PCR or long reads for confirmation.
  • Copy number and structural variants. Deletions of whole exons need depth-based calling (GATK gCNV, CNVnator) or split-read callers (Manta, Delly). Whole-genome sequencing of pancreatic tumors found structural rearrangement classes that exome-scale analysis simply does not resolve.4
  • Segmental duplications. SMN1/SMN2, PMS2/PMS2CL, CYP21A2, and the pseudogene-rich regions produce mapping ambiguity. Reads pile onto the wrong paralog.
  • Mosaicism. A variant present in 8% of blood cells will show as AD 55,5 and be filtered as noise. In retinoblastoma, a substantial share of apparently sporadic cases carry mosaic RB1 variants that require targeted deep sequencing to detect.5
  • Regulatory and deep intronic variants, which are called but not interpretable at scale.

This is one reason genome-scale testing outperforms targeted panels in undiagnosed cases. In a pediatric cohort, whole-genome sequencing raised diagnostic yield well above prior targeted testing and changed clinical management in a meaningful fraction of the diagnosed patients.6 Coverage breadth beats depth of prior assumption.

Also note what a consumer genotyping array is. A 650k-probe SNP array interrogates preselected sites, mostly common polymorphisms catalogued since the first large SNP maps.7 It cannot see a variant that has no probe, and its per-call error rate on rare alleles is high enough that any actionable result from raw array data should be confirmed by sequencing before anyone acts on it. Methylation arrays have a parallel limitation: the EPIC BeadChip covers a large but specific probe set, and probe-level performance varies enough that whole-genome bisulfite sequencing remains the comparator for anything unexpected.8

Anything that touches a medical decision (a pathogenic variant in a cancer predisposition gene, a pharmacogenomic star allele, a cardiomyopathy variant, reproductive planning) belongs with a clinical geneticist or genetic counselor working from a CLIA-validated result. Research-grade VCFs are for understanding, not for deciding.

Questions people also ask

Do you inherit 50% of your DNA from each parent? For the nuclear autosomes, yes: one copy of each of chromosomes 1–22 from each parent. Sex chromosomes are not symmetric, and mitochondrial DNA is inherited maternally, so the whole-genome share is slightly off 50/50. Sibling sharing averages 50% but varies (roughly 38–61%) because recombination shuffles which segments you get.

What does a negative DNA test result look like? Typically a line reading “no pathogenic or likely pathogenic variants detected in the genes analyzed,” followed by a gene list, a transcript list, and a technical limitations paragraph. Read the limitations section. It tells you which variant classes the assay cannot detect, and that is the real content of a negative.

What does 5 mean in DNA? It refers to the 5′ (five-prime) end of a strand, named for the carbon in the deoxyribose sugar that carries the terminal phosphate. The other end is 3′, where the free hydroxyl sits.

Do you always read DNA 5′ to 3′? Sequences are written 5′ to 3′ by convention, and polymerases synthesize in that direction. Variant coordinates in a VCF are on the plus strand of the reference, so for a gene on the minus strand the HGVS c. notation and the VCF REF/ALT will look like complementary bases. That mismatch is expected, not an error.

What do the numbers on DNA test results mean? In a clinical VCF they are quality and depth metrics (DP, GQ, AD) plus genomic and transcript coordinates. In a relationship test they are something different: allele repeat counts at STR loci, combined into a likelihood ratio such as a combined paternity index, with a probability derived from that ratio and a prior. Forensic and relationship testing runs on multiplex STR panels rather than sequence variants, and the numbers are statistical weights of evidence.9

What does 23 mean in DNA? Humans have 23 pairs of chromosomes: 22 autosome pairs plus one sex chromosome pair, 46 total in a typical somatic cell.

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. Masao Nagasaki, Jun Yasuda, Fumiki Katsuoka, et al. Rare variant discovery by deep whole-genome sequencing of 1,070 Japanese individuals. Nature Communications, 2015. https://doi.org/10.1038/ncomms9018 ↩

  2. Karen Binkley, Alvin E. Davis. Clinical, biochemical, and genetic characterization of a novel estrogen-dependent inherited form of angioedema. Journal of Allergy and Clinical Immunology, 2000. https://doi.org/10.1067/mai.2000.108106 ↩

  3. G.R. Sutherland, Robert I. Richards. Simple tandem DNA repeats and human genetic disease. Proceedings of the National Academy of Sciences, 1995. https://doi.org/10.1073/pnas.92.9.3636 ↩

  4. Australian Pancreatic Cancer Genome Initiative, Nicola Waddell, Marina Pajic, et al. Whole genomes redefine the mutational landscape of pancreatic cancer. Nature, 2015. https://doi.org/10.1038/nature14169 ↩

  5. Sameh E. Soliman, Hilary Racher, Chengyue Zhang, et al. Genetics and Molecular Diagnostics in Retinoblastoma — An Update. Asia-Pacific Journal of Ophthalmology, 2017. https://doi.org/10.22608/apo.201711 ↩

  6. Dimitri J. Stavropoulos, Daniele Merico, Rebekah Jobling, et al. Whole-genome sequencing expands diagnostic utility and improves clinical management in paediatric medicine. npj Genomic Medicine, 2016. https://doi.org/10.1038/npjgenmed.2015.12 ↩

  7. The International SNP Map Working Group, Cold Spring Harbor Laboratories:, Ravi Sachidanandam, et al. A map of human genome sequence variation containing 1.42 million single nucleotide polymorphisms. Nature, 2001. https://doi.org/10.1038/35057149 ↩

  8. Ruth Pidsley, Elena Zotenko, Timothy J. Peters, et al. Critical evaluation of the Illumina MethylationEPIC BeadChip microarray for whole-genome DNA methylation profiling. Genome biology, 2016. https://doi.org/10.1186/s13059-016-1066-1 ↩

  9. Penelope R. Haddrill. Developments in forensic DNA analysis. Emerging Topics in Life Sciences, 2021. https://doi.org/10.1042/etls20200304 ↩