How to Read Your Genome Results

Your genome results are a list of the positions where your DNA differs from a reference sequence, plus a claim about how confident the sequencer and the variant caller are in each difference, plus (in a clinical report) an interpretation of a few dozen of those differences against curated disease evidence. A whole-genome sequence at 30x coverage will produce roughly 4 to 5 million variants relative to GRCh38, of which about 3.5 million are single-nucleotide variants and several hundred thousand are short insertions and deletions. The overwhelming majority are common polymorphisms shared with millions of other people and carry no individual interpretation. Reading your results means knowing which handful of the four million are worth attention, and understanding that “positive” in a genetic report is a statement about evidence in a database, not a statement about your future.

What files you should have, and what each one is for

The deliverable that matters is not a PDF. Ask for four things: the raw reads, the alignment, the small-variant calls, and the structural-variant calls.

Raw reads arrive as FASTQ, usually two gzipped files per sample for paired-end sequencing (sample_R1.fastq.gz, sample_R2.fastq.gz), roughly 40 to 60 GB compressed for a 30x human genome. These are the only truly primary data. Everything downstream is a reprocessing decision you can redo later with a better aligner or a newer reference build, which is the reason to insist on them.

The alignment is a CRAM or BAM file with its index (.crai or .bai). CRAM is reference-compressed and runs about 15 to 20 GB against GRCh38 versus 80 to 100 GB for the equivalent BAM, so take CRAM and keep a copy of the exact reference FASTA it was compressed against. Without that FASTA, the CRAM is not decodable.

Small variants arrive as a VCF or its binary form BCF, typically gzipped and Tabix-indexed (.vcf.gz plus .vcf.gz.tbi). This is where you will spend your time. A single-sample germline VCF from GATK HaplotypeCaller or DeepVariant is a few hundred megabytes and contains one line per variant site with genotype (GT), read depth (DP), allele depths (AD), and genotype quality (GQ).

Structural variants, meaning deletions, duplications, inversions, and translocations larger than about 50 base pairs, come in a separate VCF from a caller like Manta, Delly, or (for long reads) Sniffles. Short-read structural-variant calling is the weakest part of a standard pipeline, and its false-positive rate on small deletions in repetitive regions is high enough that any interesting call needs visual confirmation in the alignment before you take it seriously.

The first hour of analysis

Before interpreting anything, check that the data are what they claim to be. Three commands get you most of the way.

# Coverage distribution and duplicate rate
samtools stats sample.cram --reference GRCh38.fa > stats.txt
mosdepth --by 1000 --fast-mode sample sample.cram

# Variant-level sanity check
bcftools stats sample.vcf.gz > vcfstats.txt

What you want to see: mean autosomal coverage at or above the promised depth (30x nominal usually means 28 to 32x aligned, non-duplicate), at least 90 percent of the callable genome at 20x or better, a PCR-duplicate rate under 10 percent, and an SNV transition-to-transversion ratio near 2.0 to 2.1 for a whole genome. A Ti/Tv well below 2.0 suggests a sea of false-positive calls. Heterozygous-to-homozygous ratios that swing far from the 1.4 to 1.6 range on autosomes indicate contamination or sample mix-up, and VerifyBamID2 will quantify that directly.

Then check the reference build. GRCh37 and GRCh38 coordinates differ, and a variant looked up at the wrong build lands in the wrong gene. Look at the VCF header for ##reference and the contig naming convention (chr1 versus 1). Mixing conventions is the single most common reason an annotation run silently returns nothing.

Annotation: turning coordinates into meaning

A raw VCF line tells you a base changed. Annotation tells you which gene and transcript it falls in, what it does to the protein, how common it is in reference populations, and whether anyone has curated it. We would run Ensembl VEP with a local cache rather than a web service, both for speed and because you should not be uploading your genotypes to someone else’s server.

vep -i sample.vcf.gz -o annotated.vcf.gz \
  --cache --offline --assembly GRCh38 \
  --vcf --compress_output bgzip \
  --everything --pick_allele_gene \
  --plugin CADD,whole_genome_SNVs.tsv.gz \
  --custom clinvar.vcf.gz,ClinVar,vcf,exact,0,CLNSIG,CLNREVSTAT

The --pick_allele_gene flag is a judgment call: it collapses the many transcript consequences per variant to one per gene, which makes the output tractable but hides the case where a variant is benign in the dominant transcript and damaging in a minor one. For a first pass, take the simplification. For any variant you are going to act on, go back and read all transcripts.

The filter that does the most work is allele frequency. Query gnomAD (v4 covers about 730,000 exomes and 76,000 genomes) and drop anything with a population maximum frequency above roughly 0.1 percent when looking for rare dominant disease variants. That single step takes 4.5 million variants to a few thousand. The reason it works is that the common variation in your genome was catalogued long ago across populations, and a variant carried by one person in fifty is not the explanation for a rare severe condition.¹ Combine frequency filtering with predicted consequence (nonsense, frameshift, canonical splice site, and missense with high CADD) and gene-level relevance, and you are looking at a list a person can read.

What a “positive” result means, and what it does not

Clinical labs classify variants on the five-tier ACMG/AMP scale: pathogenic, likely pathogenic, uncertain significance, likely benign, benign. A positive report means one variant reached pathogenic or likely pathogenic in a gene with an established disease association. Three things follow from how that classification is built.

First, the classification describes a variant, not a person. Penetrance, meaning the fraction of carriers who develop the condition, varies enormously by gene and by how the carriers were ascertained. Estimates derived from families with many affected relatives are systematically higher than estimates from unselected population cohorts, because the family also carries modifier variants and shared environment. A pathogenic variant found incidentally in a healthy person with no family history usually carries lower risk than the textbook number.

Second, classifications change. ClinVar entries are revised as evidence accumulates, and a variant of uncertain significance today may be reclassified in either direction in five years. This is why holding the primary data matters: you can re-annotate the same VCF against next year’s database in an afternoon.

Third, ancestry biases the whole system. Reference variant databases and the evidence behind curated classifications draw disproportionately on European-ancestry cohorts, which inflates the rate of uncertain results for everyone else. Careful frequency comparison against an ancestry-matched reference population is the correction, and population-scale variation maps exist precisely to make that comparison possible.¹

The categories of condition that genome sequencing detects well are rare Mendelian disorders caused by single high-impact variants, carrier status for recessive conditions relevant to reproductive planning, dominant cancer-predisposition syndromes in genes such as BRCA1, BRCA2, the Lynch syndrome mismatch-repair genes, and TP53, cardiac channelopathies and cardiomyopathies, and pharmacogenomic genotypes in genes including CYP2C19, CYP2D6, DPYD, and TPMT. Sequencing detects common complex disease poorly through any single variant, because the effect is distributed across thousands of loci. And germline sequencing says nothing about the somatic mutations that accumulate in a tumour, which are a separate measurement on separate tissue.²

If a variant in your results is classified pathogenic or likely pathogenic in a gene with clinical management implications, take it to a board-certified clinical geneticist or genetic counselor before drawing any conclusion, and have it confirmed on a clinical-grade assay. Research-grade and consumer pipelines produce false positives, particularly in repetitive regions, in genes with pseudogenes, and for indels near homopolymers. Confirmation by an orthogonal method in an accredited lab is the standard step, and interpretation in the context of your personal and family history is medical work that belongs to a clinician.

Where a genome stops

The reference genome you are compared against was finished in 2004 at about 99 percent of the euchromatic sequence with an error rate near one per 100,000 bases, and the residual gaps sat in exactly the segmental duplications and satellite arrays where short-read calling still struggles.³ The practical consequence is that a small fraction of your genome is not confidently callable with 150 bp reads, and long-read sequencing is the answer where a region matters.

The deeper limit is that sequence is static. Your genome is identical at 20 and at 60, while your expression, protein levels, and metabolic state are not. The promise that came with the Human Genome Project was that sequence would reorganize medicine around individual biology, and much of what has been delivered is prediction of risk rather than measurement of current state.⁴ Reading your genome results well means treating them as the constant term: the layer that constrains what is possible, interpreted alongside measurements that change.

Questions people also ask

What happens if genetic testing comes back positive? The lab reports a pathogenic or likely pathogenic variant in a specific gene, and the next steps are confirmation on a clinical assay and a consultation with a genetic counselor or clinical geneticist who will interpret it against your personal and family history. Positive means evidence supports the variant causing disease in carriers generally, not that you have or will develop the condition.

What diseases can be detected by genetic testing? Rare single-gene disorders, recessive carrier states, dominant cancer-predisposition syndromes, inherited cardiac conditions, and drug-metabolism genotypes are all well covered. Common multifactorial disease is not detectable from any single variant, and tumour mutations require sequencing the tumour rather than blood.²

How do I read my genetic test results? On a clinical report, read the gene, the variant in HGVS notation (for example NM_007294.4:c.5266dupC), the zygosity, and the ACMG classification, in that order. On raw data, annotate the VCF with VEP, filter to population frequency under 0.1 percent, and review high-impact consequences in genes with established disease links.

What does an abnormal genetic test mean? Everyone carries millions of differences from the reference, so “abnormal” is meaningful only when a specific variant is rare and has curated evidence tying it to a phenotype.¹ A variant of uncertain significance is a common and genuinely inconclusive outcome, and it should not be treated as a finding until reclassified.

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. The 1000 Genomes Project Consortium. A map of human genome variation from population-scale sequencing. Nature, 2010. https://doi.org/10.1038/nature09534 ↩ ↩² ↩³

2. The Cancer Genome Atlas Network. Comprehensive molecular portraits of human breast tumours. Nature, 2012. https://doi.org/10.1038/nature11412 ↩ ↩²

3. International Human Genome Sequencing Consortium. Finishing the euchromatic sequence of the human genome. Nature, 2004. https://doi.org/10.1038/nature03001 ↩

4. Francis S. Collins. Medical and Societal Consequences of the Human Genome Project. New England Journal of Medicine, 1999. https://doi.org/10.1056/nejm199907013410106 ↩