What a Genome Testing Kit Can and Cannot Tell You

A genome testing kit sent to your house can sequence your DNA well enough to answer a large class of questions: ancestry, carrier status for recessive conditions with well-characterized variants, pharmacogenetic star alleles, and the presence of specific pathogenic variants in genes like BRCA1, LDLR, or HFE. Whether you have a genetic disease is a determination that requires a clinician, usually a confirmatory test in a CLIA-certified laboratory, and a phenotype to interpret the variant against. A kit cannot, on its own, tell you that. The useful way to think about a kit is as a data acquisition step whose value depends almost entirely on which assay you bought and whether you receive the raw files.

The distinction that matters most is array versus sequencing. A genotyping array (the technology behind most sub-$200 kits) interrogates a fixed panel of roughly 600,000 to 900,000 predetermined positions using allele-specific probes. It reports a call at each of those positions and says nothing about the 3.1 billion bases in between. Whole-genome sequencing reads the genome by synthesis at some average depth, typically 30x for a consumer product, and can in principle call a variant anywhere the reads align. These produce fundamentally different data objects, and the difference shows up immediately when you try to answer a question the array designers did not anticipate.

What each assay measures

An array returns genotypes at sites chosen years ago for allele frequency and tagging efficiency in reference panels derived from projects like the 1000 Genomes Project, which mapped common variation across populations and established the haplotype scaffolds that imputation depends on.¹ This design choice has a direct consequence: arrays are excellent at common variants (minor allele frequency above roughly 1 percent) and poor at rare ones. Most pathogenic variants in Mendelian disease genes are rare. A typical array carries a few thousand hand-picked clinically relevant probes, so it will detect the three Ashkenazi founder variants in BRCA1 and BRCA2 and miss the several thousand other pathogenic variants in those genes.

Array data also suffer from a specific and underappreciated failure mode: false positives at rare sites. Probe intensity clustering is calibrated on common genotypes, and when a rare variant is present the clustering algorithm frequently assigns the wrong call. Published evaluations of direct-to-consumer array data have found false positive rates for rare pathogenic variants high enough that no result of that kind should be treated as real without orthogonal confirmation by Sanger sequencing or a targeted clinical assay.

Whole-genome sequencing at 30x mean depth calls single nucleotide variants and small insertions and deletions across the accessible genome with high sensitivity, typically above 99 percent for SNVs in high-confidence regions. It also captures copy number changes, structural rearrangements, and repeat expansions, though each of those requires a different caller and each degrades with short reads. Sequencing-based approaches routinely find variation that fixed-panel assays cannot see, a pattern documented across domains: next-generation sequencing of clinical samples detects variants and organisms that commercial kits with predetermined probe sets simply do not report.² The same logic applies to your genome.

Insist on raw files, and know which ones

If you are technically fluent and intend to analyze your own data, the deliverable is the point. Ask before you buy which of the following you receive:

• FASTQ (gzipped), the raw reads with per-base Phred quality scores. Around 100 GB for a 30x human genome. This is the only format that lets you realign to a different reference or rerun the entire pipeline.
• CRAM or BAM aligned to a stated reference. Insist on GRCh38 with alt contigs, ideally the analysis set used by the Genome Analysis Toolkit. CRAM is roughly 15 to 20 GB at 30x, BAM closer to 60 GB. If a vendor only offers hg19, you will spend real time on liftover and will lose accuracy in regions where the assembly changed.
• gVCF or VCF with per-sample genotype quality (GQ), depth (DP), and allelic depth (AD) fields populated. A VCF that reports only genotypes without quality annotations is nearly useless for filtering.

A vendor that returns only a polished web report and a TSV of interpreted variants has sold you an interpretation. You did not buy data. You cannot check their work, rerun with a newer caller, or ask a question they did not anticipate.

A pipeline we would run

Start by verifying what you were sent. Run samtools flagstat and samtools stats on the CRAM and confirm the mapping rate is above 98 percent and the duplicate rate is below roughly 15 percent. Run mosdepth --by 1000 to get coverage distribution, and check that mean depth matches the advertised number and that the fraction of the genome at 10x or better exceeds 90 percent. Coverage dropout in GC-rich first exons is common and directly limits which genes you can interpret.

For variant calling, we would use DeepVariant rather than GATK HaplotypeCaller for germline short variants on a single sample. DeepVariant’s convolutional model over pileup images gives slightly better indel precision on Illumina data and requires almost no parameter tuning. Run it with --model_type=WGS and evaluate the output against the Genome in a Bottle truth sets using hap.py restricted to the high-confidence BED. If you are not hitting F1 above 0.995 for SNVs, something upstream is wrong.

For annotation, use Ensembl VEP with --everything --assembly GRCh38, plus plugins for CADD and SpliceAI, and join against ClinVar with bcftools annotate. Filter to variants with a ClinVar assertion of pathogenic or likely pathogenic and at least two submitters, then look at the genes returned. In a randomly chosen person this list is short, often a handful of carrier-state findings in recessive genes.

Two caveats that will bite you. First, pharmacogenetics requires phased star-allele calling, not per-variant genotypes: run PharmCAT or Aldy on the aligned reads, because CYP2D6 in particular involves copy number variation and hybrid genes that no VCF line will describe. Second, the mitochondrial genome is a separate problem. It is present at hundreds of copies per cell, heteroplasmic, and needs a caller that reports variant allele fraction rather than a diploid genotype. Dedicated mitochondrial sequencing methods exist precisely because standard workflows handle it badly.³

What sequencing still misses

Short-read whole-genome sequencing is blind or nearly blind to several variant classes. Repeat expansions in genes like HTT, FMR1, and the C9orf72 hexanucleotide are larger than the read length, and while ExpansionHunter can estimate sizes from anchored read pairs, it is not a substitute for a clinical repeat-primed PCR. Segmental duplications such as the SMN1/SMN2 region confound alignment, so carrier screening for spinal muscular atrophy needs a targeted assay like SMNCopyNumberCaller. Deep intronic and regulatory variants are called but rarely interpretable. And many negative results stay negative because the gene-disease link has not been established yet, which is why functional work in model organisms remains part of solving rare disease cases.⁴

Sequencing also does not replace clinical context. In settings where a specific population carries known founder variants, targeted screening with a short defined panel outperforms a broad untargeted scan on both cost and actionability.⁵ Whole-genome sequencing earns its keep when the question is open-ended or when a single assay needs to replace many, as in pediatric leukemia where one genome replaces karyotyping, FISH, and fusion panels simultaneously.⁶

Questions people also ask

Can I test for genetic diseases at home?

You can collect a sample at home and have your genome sequenced, and the data will contain variants relevant to genetic disease. Interpreting that data as a diagnosis requires a clinician. In nearly all cases it also requires confirmation of the specific variant in a clinical laboratory. The reason is that consumer pipelines are not validated for diagnostic use, and rare-variant calls from arrays in particular have substantial false positive rates.

What does genome testing test for?

Whole-genome sequencing detects single nucleotide variants, small insertions and deletions, copy number changes, and some structural variants across the genome. From those you can derive carrier status, pharmacogenetic haplotypes, polygenic scores, ancestry, and the presence or absence of specific known pathogenic variants. Arrays test only the fixed set of positions the manufacturer chose.

What can genome testing tell you that a blood panel cannot?

The genome is fixed and gives you risk and carrier information that does not change over your lifetime, along with pharmacogenetic haplotypes relevant to drug metabolism. Genome data tells you nothing about your current physiological state. Claims that genotype should direct specific dietary changes remain, for most variants, poorly supported by outcome evidence.⁷

Do I need the raw data, or is a report enough?

If you want to ask questions beyond the vendor’s report, you need FASTQ or CRAM plus an annotated VCF. Reports age badly as variant classifications are revised in ClinVar, while raw reads can be reanalyzed indefinitely.

Footnotes

1. Elizabeth Pennisi. 1000 Genomes Project Gives New Map of Genetic Diversity. Science, 2010. https://doi.org/10.1126/science.330.6004.574 ↩

2. Tracy L Meiring, Anna T Salimo, Beatrix Coetzee, et al. Next-generation sequencing of cervical DNA detects human papillomavirus types not detected by commercial kits. Virology Journal, 2012. https://doi.org/10.1186/1743-422x-9-164 ↩

3. Yue Yao, Motoi Nishimura, Kei Murayama, et al. A simple method for sequencing the whole human mitochondrial genome directly from samples and its application to genetic testing. Scientific Reports, 2019. https://doi.org/10.1038/s41598-019-53449-y ↩

4. Shinya Yamamoto, Oguz Kanca, Michael F. Wangler, et al. Integrating non-mammalian model organisms in the diagnosis of rare genetic diseases in humans. Nature Reviews Genetics, 2023. https://doi.org/10.1038/s41576-023-00633-6 ↩

5. Kevin A. Strauss, Erik G. Puffenberger, D. Holmes Morton. One Community’s Effort to Control Genetic Disease. American Journal of Public Health, 2012. https://doi.org/10.2105/ajph.2011.300569 ↩

6. Sarra L. Ryan, John F. Peden, Zoya Kingsbury, et al. Whole genome sequencing provides comprehensive genetic testing in childhood B-cell acute lymphoblastic leukaemia. Leukemia, 2023. https://doi.org/10.1038/s41375-022-01806-8 ↩

7. Kathryn M. Camp, Elaine Trujillo. Position of the Academy of Nutrition and Dietetics: Nutritional Genomics. Journal of the Academy of Nutrition and Dietetics, 2014. https://doi.org/10.1016/j.jand.2013.12.001 ↩