How to Interpret Your Own Genetic Test Results

By the end of this guide you will be able to take either a clinical genetic report or a raw variant file and answer three questions about any single finding: what change was observed, how confident the sequencing is that the change is real, and what the evidence says about whether that change matters. You need a VCF (Variant Call Format, the standard text file of variant calls) with its index. Ideally you also have the aligned reads it came from as CRAM or BAM so you can inspect calls directly. You also need a reference genome FASTA matching the build used for alignment and a Linux or macOS machine with about 100 GB of free disk. Tools: bcftools 1.19+, samtools, Ensembl VEP (Variant Effect Predictor) or ANNOVAR, tabix, and IGV for visual inspection. Everything below assumes GRCh38 unless stated. If your data is GRCh37, do not mix annotation sources across builds, because coordinates differ by millions of bases in places and you will silently annotate the wrong gene.

1. Establish what kind of test produced the result

Before reading a single variant, determine the assay, because the assay determines what could not have been found. A targeted panel interrogates a fixed gene list. A negative result says nothing about genes outside it. Exome sequencing covers roughly 1–2% of the genome and misses deep intronic and most structural variation. Whole-genome sequencing covers the rest at uniform depth but still calls short variants far better than it calls large rearrangements and repeat expansions. Genotyping arrays, the technology behind most consumer DNA products, probe a few hundred thousand to a million pre-chosen positions. They do not sequence at all, so an array “negative” for a gene is close to meaningless.

This distinction is the single most common source of misreading. The three classical result categories in clinical genetics are positive, negative, and a variant of uncertain significance, or VUS. Positive means a pathogenic or likely pathogenic variant was found. Negative means no such variant was found in what was tested. A negative is always conditional on the assay’s reach and on the reference against which the sample was compared, and the interpretive standards for whole-genome testing explicitly ask laboratories to state coverage and technical limitations alongside the result ¹. If your report does not state the gene list and the fraction of target covered at 20× or better, treat any negative as provisional.

2. Parse the variant nomenclature

A report will name a variant in HGVS (Human Genome Variation Society) notation, which looks like NM_000059.4:c.5946delT (p.Ser1982Argfs*22). Read it in pieces. The prefix is the reference sequence. An NM_ prefix is a specific RefSeq transcript, and the version suffix matters because exon numbering and coordinates shift between versions. The c. prefix denotes a coding-DNA position, while g. denotes a genomic position and n. a non-coding transcript position. The 5946delT term is the change itself. The parenthetical p. term is the predicted protein consequence, with fs*22 meaning a frameshift that reaches a stop codon 22 residues downstream.

Two practical consequences follow. First, the protein consequence is a prediction from one transcript, and a variant that truncates the canonical transcript may be harmless in a tissue that predominantly expresses a shorter isoform. Always check which transcript the laboratory used against the MANE Select transcript (Matched Annotation from NCBI and EMBL-EBI) for that gene. Second, nomenclature is not the same thing as genomic position, so to look a variant up anywhere else you need to convert it. bcftools will normalize and left-align representations so that the same biological change gets one canonical form:

bcftools norm -f GRCh38.fa -m -any -c w \
  -Oz -o sample.norm.vcf.gz sample.vcf.gz
bcftools index -t sample.norm.vcf.gz

Normalization matters more than it sounds. An indel in a homopolymer can be written several equivalent ways, and an un-normalized query against gnomAD or ClinVar will return no match for a variant that is present in both.

3. Annotate the file yourself

Run your own annotation rather than trusting a vendor’s spreadsheet. You then control which transcript set, which population reference, and which version of ClinVar you compared against. VEP with a local cache is the approach we use, since it is deterministic, offline, and reports consequences per transcript:

vep --cache --offline --assembly GRCh38 \
  --fasta GRCh38.fa \
  --input_file sample.norm.vcf.gz --format vcf \
  --vcf --output_file sample.vep.vcf.gz --compress_output bgzip \
  --everything --pick_allele_gene --mane_select \
  --custom file=clinvar_20260601.vcf.gz,short_name=ClinVar,format=vcf,type=exact,fields=CLNSIG%CLNREVSTAT \
  --custom file=gnomad.genomes.v4.1.sites.vcf.gz,short_name=gnomADg,format=vcf,type=exact,fields=AF%AF_popmax \
  --fork 8

--pick_allele_gene gives one consequence per gene per allele, which keeps the output readable, but keep --everything so the per-transcript detail is still in the CSQ field when you need it. --mane_select flags the transcript clinical laboratories are most likely to have used. The two --custom blocks are where the interpretive signal lives: ClinVar’s assertion plus its review status, and population allele frequency.

A whole genome yields roughly four to five million variants per person relative to the reference, and the great majority are common and benign ². Filtering is therefore the substance of interpretation rather than a preliminary step. Start by restricting to rare, protein-altering, and previously asserted variants:

bcftools +split-vep sample.vep.vcf.gz \
  -c SYMBOL,Consequence,MANE_SELECT,ClinVar_CLNSIG,ClinVar_CLNREVSTAT,gnomADg_AF:Float \
  -s worst -Ou \
| bcftools view -i 'FILTER="PASS" && (gnomADg_AF<0.001 || gnomADg_AF=".") && (ClinVar_CLNSIG~"athogenic" || Consequence~"frameshift|stop_gained|splice_acceptor|splice_donor|start_lost")' \
  -Oz -o candidates.vcf.gz

Expect on the order of a few hundred rows. That is a list of things worth looking at. It is not a list of problems.

4. Read the classification, and read the evidence behind it

Clinical laboratories classify variants on a five-tier scale. The tiers are pathogenic, likely pathogenic, uncertain significance, likely benign and benign. “Likely” in this system carries a conventional quantitative meaning of roughly 90% confidence, and the tiers are assigned by weighing evidence lines: population frequency, functional assays, segregation in families, computational prediction, and case-control data. Recommendations for structured, tiered reporting of sequence variants in susceptibility genes predate the current standards and were developed precisely because free-text laboratory opinions were not comparable across labs ³.

Two things follow for you as a reader. The first is that a classification is a summary of evidence at a point in time and will change. Reclassification is routine, most often from VUS toward benign as population databases grow. The second is that laboratories reach different conclusions on the same variant. A survey of interpretation and reporting practices across US laboratories found substantial variation in the evidence sources used, the thresholds applied, and what is returned to the ordering clinician ⁴. When ClinVar shows conflicting assertions for your variant, look at the review status. An expert-panel submission from ClinGen’s gene-specific panel, for instance, carries much more weight than a single submitter with no assertion criteria provided.

For a VUS, the operationally correct interpretation is that the variant is not to be used for medical decisions. A VUS is not a weak positive. Treating it as one is the most damaging error in this area, and it is what the literature on interpretive pitfalls keeps returning to ⁵.

5. Check the sequencing before you believe the biology

Every classification above assumes the call is real. Verify that yourself for anything you care about. Pull depth and allele balance at the site:

bcftools query -f '%CHROM\t%POS\t%REF\t%ALT\t%QUAL\t[%GT\t%DP\t%AD]\n' \
  -r chr13:32338763 sample.norm.vcf.gz

For a true heterozygous single-nucleotide variant you want depth of at least 20×, an alternate allele fraction near 0.5 (roughly 0.3 to 0.7 is normal sampling noise), and a genotype quality above 20. Allele fractions near 0.2 in a germline call usually indicate a mapping artifact or a nearby indel confusing the aligner. Then look at the reads. Load the CRAM in IGV and navigate to the position. Check whether alternate reads come from both strands, whether they cluster at read ends, and whether the region is full of soft-clipped reads.

Regions of the genome differ enormously in how reliably they can be sequenced and mapped. Segmental duplications, high-GC first exons, and paralogous gene families such as PMS2/PMS2CL or SMN1/SMN2 produce both false positives and false negatives with short reads. Design and interpretation of sequencing studies has to account for this non-uniformity explicitly, and the same logic applies to a single genome read by one person ⁶. A practical check on your own file:

samtools coverage -r chr7:117,480,025-117,668,665 sample.cram   # CFTR
mosdepth --by exons.bed --fast-mode sample sample.cram

If mean depth over a gene of interest falls below about 15× or callable fraction below 90%, you have not tested that gene, whatever your VCF says.

6. Translate a confirmed variant into risk, carefully

A pathogenic variant is a statement about the variant. It is not a prediction about you. Penetrance, the probability that a carrier develops the associated condition, varies from near-complete to under 10% depending on the gene, the specific variant, sex, and genetic background. Penetrance estimates derived from clinically ascertained families, where testing happened because several relatives were affected, are systematically higher than estimates from unselected population cohorts. When you read a risk figure, ask which population it came from.

Inheritance mode changes the meaning of a single finding entirely. One pathogenic variant in an autosomal recessive gene makes you a carrier with typically no personal risk, though it matters for reproductive planning. For autosomal dominant conditions a single variant is potentially consequential. For a recessive condition you need to know whether two variants are on the same chromosome copy (in cis, biologically equivalent to one hit) or on opposite copies (in trans, a functional loss of both alleles). Phasing from short reads is only reliable when variants are within a read-pair span, so this question often requires testing a parent.

Polygenic scores are a separate category and should not be read the way you read a monogenic result. A score places you on a distribution relative to the cohort that trained it, its transferability across ancestries is limited, and it does not identify any causal variant in you. Clinical genome interpretation practice keeps monogenic findings, carrier status, pharmacogenomic genotypes, and polygenic estimates in distinct reporting sections for exactly this reason ¹. In cancer genomics, where somatic tumor variants and inherited germline variants arrive on the same report and have completely different implications, the distinction between the two is the first thing a reader has to get right ⁷.

7. Decide what requires a clinician

Some conclusions you can reach alone and some you cannot. You can reasonably confirm a call technically and find the gene’s inheritance pattern. You can also read ClinVar’s review status and understand what your assay did and did not cover. You cannot self-manage a pathogenic finding in a medically actionable gene, and you should not try. Some findings require a genetics professional and, in most jurisdictions, confirmation in a clinically accredited laboratory using an orthogonal method. That applies to a confirmed or suspected pathogenic variant in a cancer predisposition gene, a cardiomyopathy or arrhythmia gene, or a familial hypercholesterolemia gene. It also applies to any finding you are considering acting on. Research-grade sequencing is not diagnostic, and the step from a research VCF to a clinical conclusion is a formal process for good reasons ⁸.

Never start, stop, or change a medication on the basis of a genotype you interpreted yourself, including pharmacogenomic results. That is a clinical decision.

Common problems

Build mismatch is the most frequent and the most silent. A GRCh37 VCF annotated against a GRCh38 cache produces plausible-looking output with wrong genes. Check the header (bcftools view -h | grep -i assembly) and confirm a known position: TP53 sits near chr17:7,676,000 in GRCh38 and chr17:7,577,000 in GRCh37.

Un-normalized representation causes false negatives in every lookup. Always run bcftools norm -f before querying ClinVar or gnomAD, and remember that multi-allelic sites must be split with -m -any or filters will behave unpredictably.

Transcript choice changes the answer. A variant labeled stop_gained on a minor transcript may be 3_prime_UTR_variant on MANE Select. Filter on the MANE consequence and check the alternatives before concluding anything.

Population frequency without ancestry context misleads. A variant at 0.5% global frequency may be at 4% in one population group and genuinely benign there while appearing rare in aggregate. Use population-specific allele frequencies, and note that non-European ancestries remain underrepresented in reference databases, which inflates VUS rates for those individuals.

Reference-based variant calling systematically misses what is not in the reference. Repeat expansions, large deletions, and inversions need dedicated callers (ExpansionHunter, manta, dragen-sv) and confirmatory assays. A short-variant VCF being clean says nothing about these classes ⁶.

Absence of evidence gets read as evidence of absence. Roughly one to two percent of people carry a pathogenic variant in a well-characterized actionable gene, so a clean genome is the common outcome and does not lower your baseline population risk for anything. Family history remains independently informative and often more predictive than a negative sequencing result ⁵.

Version drift breaks reproducibility. Record the exact versions of your reference FASTA, VEP cache, ClinVar release, and gnomAD release alongside your results, and re-run your candidate list annually. Variants get reclassified, and a finding you dismissed as uncertain in one year may have an expert-panel assertion the next.

Footnotes

1. Christina A. Austin-Tse, Vaidehi Jobanputra, Denise L. Perry, et al. Best practices for the interpretation and reporting of clinical whole genome sequencing. npj Genomic Medicine, 2022. https://doi.org/10.1038/s41525-022-00295-z ↩ ↩²

2. Rachel L Goldfeder, Dennis P Wall, Muin J Khoury, et al. Human Genome Sequencing at the Population Scale: A Primer on High-Throughput DNA Sequencing and Analysis. American Journal of Epidemiology, 2017. https://doi.org/10.1093/aje/kww224 ↩

3. Sharon E. Plon, Diana M. Eccles, Douglas Easton, et al. Sequence variant classification and reporting: recommendations for improving the interpretation of cancer susceptibility genetic test results. Human Mutation, 2008. https://doi.org/10.1002/humu.20880 ↩

4. Julianne M. O’Daniel, Heather M. McLaughlin, Laura M. Amendola, et al. A survey of current practices for genomic sequencing test interpretation and reporting processes in US laboratories. Genetics in Medicine, 2017. https://doi.org/10.1038/gim.2016.152 ↩

5. Hubert E Blum, Konrad Oexle. Clinical Interpretation and Implications of Whole Genome Sequencing. Journal of Next Generation Sequencing & Applications, 2014. https://doi.org/10.4172/2469-9853.1000105 ↩ ↩²

6. David B. Goldstein, Andrew Allen, Jonathan Keebler, et al. Sequencing studies in human genetics: design and interpretation. Nature Reviews Genetics, 2013. https://doi.org/10.1038/nrg3455 ↩ ↩²

7. Raffaella Casolino, Philip A. Beer, Debyani Chakravarty, et al. Interpreting and integrating genomic tests results in clinical cancer care: Overview and practical guidance. CA: A Cancer Journal for Clinicians, 2024. https://doi.org/10.3322/caac.21825 ↩

8. A.J. Marian. Sequencing Your Genome: What Does it Mean?. Methodist DeBakey Cardiovascular Journal, 2014. https://doi.org/10.14797/mdcj-10-1-3 ↩