What Genome Sequencing Reviews Do and Don't Tell You
Most reviews of genome sequencing companies, including the Trustpilot and Reddit threads that rank for these searches, are reviews of customer service. They tell you whether the kit arrived, whether the turnaround matched the promise, and whether a subscription charge appeared later. They tell you almost nothing about the thing you are buying, which is a set of files with measurable properties: sequencing depth, library chemistry, alignment quality, variant call completeness, and whether you get to keep the raw reads. Sequencing.com is a real company that returns real data, and the recurring complaint in its reviews (a monthly charge attached to report access) is a business-model problem rather than a sequencing problem. Our view is that you should ignore star ratings, ask eight specific questions before paying, and then verify the delivered data yourself with standard tools.
Why review sites are the wrong instrument
A five-star review saying “the reports were in depth and gave me the answers I needed” is unfalsifiable. The customer has no second sample, no reference genome to compare against, and no way to know whether their nominal 30x whole-genome sequencing came back at 22x mean depth with a 19% duplicate rate. Reviews are also silent on the part that ages best. The interpretation of a variant changes every few years as databases grow, so the durable asset is the raw and aligned data you can reanalyze later 1.
Complaints are more informative than praise, because complaints are usually about things a customer can observe. Across consumer genomics vendors, the ones that repeat are worth taking seriously: surprise subscriptions gating access to your own results, raw data downloads that require an extra fee or a support ticket, files delivered without index files, long silences during lab processing, and sample failures that restart the clock. None of these require technical knowledge to notice, which is exactly why they dominate the reviews. The quality problems that do require technical knowledge go unreported.
The eight questions to ask before paying
Ask these in writing, before you order. A vendor that answers all eight crisply is usually fine.
- What platform and read configuration? Illumina paired-end 2x150 base pairs is the default for short-read whole-genome sequencing and is what most pipelines assume. Anything shorter degrades mapping in repetitive regions.
- Is the library PCR-free? PCR-free libraries have lower duplicate rates and better coverage uniformity in GC-rich regions, which matters for indel and copy-number calling.
- What mean depth, and what fraction of the genome at 20x or more? “30x” is a target and not a guarantee. We ask for the per-sample metric and do not settle for the assay spec.
- Is this true whole-genome sequencing or low-pass with imputation? Some products sequence at 0.4x to 1x and impute the rest against a reference panel. That is a legitimate method for common variants and a poor substitute for anything rare, since imputation cannot recover a variant absent from the panel. De novo and private variants are precisely the ones it misses 2.
- Which files do I receive, and for how long? The set we consider acceptable is FASTQ or CRAM plus index, a gVCF or VCF against a stated reference build, and a QC report. Ask which build: GRCh38 or hg38 are standard now, and older GRCh37 output means liftover work later.
- Is the laboratory CLIA-certified and CAP-accredited, or is the output labeled research use only? Research-use data is still useful for your own analysis. It usually cannot be carried into a clinical setting without a confirmatory test.
- Is there any recurring charge, and does it gate data access? This is the single most common review complaint. Get the answer in writing.
- What is the deletion policy, and can the sample be destroyed on request? Treat genomic data as permanently identifying, because it is.
Verify the delivery yourself
Once the files land, an hour of work tells you more than every review you read. Start with the alignment. If you received FASTQ, align with bwa-mem2 mem -t 16 -R '@RG\tID:1\tSM:you\tPL:ILLUMINA' or dragen-os, then mark duplicates. If you received a CRAM, confirm you also have the exact reference FASTA it was compressed against, since CRAM is reference-dependent and will not decode without it.
Then measure coverage and library quality:
samtools stats -@ 8 sample.cram > stats.txt # duplicate rate, insert size, error rate
mosdepth --by 1000 --fast-mode sample sample.cram # depth distribution, callable fraction
We look for a mean depth at or above the promised number and 90% or more of the autosomes at 20x or better. We also look for a duplicate rate under 10% and a median insert size near 350 base pairs. A duplicate rate above 20% suggests a low-input library amplified hard, which shows up as uneven coverage rather than as a missing file.
Next, confirm the sample is yours and is not contaminated:
verifybamid2 --SVDPrefix resource/1000g.phase3.100k.b38 --BamFile sample.cram
somalier extract -d extracted/ --sites sites.hg38.vcf.gz -f ref.fa sample.cram
somalier relate extracted/*.somalier
A FREEMIX estimate above roughly 0.02 indicates cross-sample contamination. Somalier also reports inferred genetic sex, which catches sample swaps quickly. If you have an older array file from a consumer ancestry test, convert it to VCF and run bcftools gtcheck against your new calls. Concordance on shared sites should be very high, and anything below about 98% means one of the two datasets is not from you.
Finally, sanity-check the variant call set:
bcftools stats -F ref.fa sample.vcf.gz > vcfstats.txt
A single human genome against GRCh38 yields on the order of four to five million small variants, with single-nucleotide substitutions outnumbering indels by roughly ten to one. A call set with 1.2 million variants was made from low-pass data or filtered aggressively. A call set with 12 million is full of artifacts. Population-scale sequencing efforts established these expectations, and they are the cheapest possible check on a pipeline 3 4.
What the reports are worth
Vendor reports are database lookups with a layer of design on top. Most pull from ClinVar, PharmGKB, and published association studies. They then filter your variants against them. The quality difference between vendors lies in curation: which submissions they trust, how they handle conflicting classifications, and whether they report variants of uncertain significance at all. That curation is often treated as proprietary, and the practice of holding variant interpretations as trade secrets has been a live controversy in the field 5.
This is why we weight raw data availability far above report polish. You can regenerate a report from your CRAM and gVCF at any time using open resources. You cannot regenerate the sequencing. The analytical challenges in personal genome interpretation have been understood for well over a decade, and they have not been solved by nicer dashboards 6. Sequencing chemistry and cost have improved enormously over that period, so the data you buy today is good 7. The interpretation remains the hard part.
Where a clinician is required
Nothing in a consumer sequencing report is a diagnosis. If a report flags a pathogenic variant in a gene with medical consequences, the next step is a genetic counselor and an orthogonal confirmatory test in a clinical laboratory, ordered by a physician. Short-read sequencing has known blind spots, including certain repeat expansions, pseudogene-adjacent regions, and structural variants, and false positives in those regions are common. We do not interpret pharmacogenomic results as treatment guidance, and neither should you without a clinician who can see your full medical picture.
Questions people also ask
Is Sequencing.com a legitimate company? Yes. It ships real kits, uses real laboratories, and returns real whole-genome data. The dominant complaint in its public reviews concerns a recurring subscription tied to report access, so read the billing terms before ordering and confirm in writing that raw data download is included.
Is whole-genome sequencing worth the money? It is worth it if you intend to work with the files. At current prices, a 30x PCR-free genome with FASTQ or CRAM delivery is a one-time purchase whose value grows as annotation databases improve 1. If you only want a curated report you will read once, the cost per insight is poor.
Which company is best for DNA sequencing? We would not name one, because the specifications matter more than the brand. Choose any vendor offering PCR-free 2x150 sequencing at 30x or greater from a CLIA-certified laboratory. It should deliver FASTQ or CRAM plus gVCF, include a per-sample QC report, and place no subscription gate on your data.
Are consumer DNA sequencing results reliable? Analytically, short-read sequencing at 30x calls common single-nucleotide variants with high accuracy, which is why it underpins population-scale research 3. Reliability drops for indels in repetitive sequence, structural variants, and anything the pipeline filters out. Clinical reliability is a separate question and depends on laboratory accreditation and confirmatory testing.
Should I trust BBB or Trustpilot ratings here? Use them only for logistics and billing signals, which is what reviewers can observe. They carry no information about depth, duplicate rate, contamination, or call set quality, and those are the properties that determine whether your data is usable.
Footnotes
-
S.M. Caspar, N. Dubacher, A.M. Kopps, et al. Clinical sequencing: From raw data to diagnosis with lifetime value. Clinical Genetics, 2018. https://doi.org/10.1111/cge.13190 ↩ ↩2
-
Joris A. Veltman, Han G. Brunner. De novo mutations in human genetic disease. Nature Reviews Genetics, 2012. https://doi.org/10.1038/nrg3241 ↩
-
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 ↩ ↩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 ↩
-
Robert Cook-Deegan, John M Conley, James P Evans, et al. The next controversy in genetic testing: clinical data as trade secrets?. European Journal of Human Genetics, 2012. https://doi.org/10.1038/ejhg.2012.217 ↩
-
Michael Snyder, Jiang Du, Mark Gerstein. Personal genome sequencing: current approaches and challenges. Genes & Development, 2010. https://doi.org/10.1101/gad.1864110 ↩
-
Jay Shendure, Shankar Balasubramanian, George M. Church, et al. DNA sequencing at 40: past, present and future. Nature, 2017. https://doi.org/10.1038/nature24286 ↩
Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.