Skip to content

Is Promethease Worth It?

Oak
A lab instrument holding a glass grid plate where only scattered wells glow blue, with a continuous glowing filament spooling behind it.

Promethease is worth roughly what it costs. For about $12 it takes a raw genotype file from 23andMe, AncestryDNA, MyHeritage or similar. It matches your genotypes against SNPedia and returns an HTML report of every annotated variant it can find. That is a useful literature index and a fast way to see what the published record says about the specific positions your array measured. It is not a health assessment. The two failure modes that matter are inherited from the input file and from SNPedia’s curation. Neither comes from Promethease itself.

If you are technically fluent and willing to spend an afternoon, you can get a better result from the same raw file using open tools. We describe that pipeline below. The summary of Promethease is that it solves the easy half of the problem (joining your genotypes to a database) and leaves the hard half (deciding what any of it means) entirely to you.

What Promethease does mechanically

The input is a tab-delimited text file with four columns. The columns are rsid, chromosome, position, and genotype. A 23andMe v5 file has roughly 640,000 rows. Promethease parses that, looks up each rsid in SNPedia, and emits an annotation for every genotype SNPedia has a page for. A typical report contains 15,000 to 25,000 entries.

Each entry carries two curated fields. Magnitude is a SNPedia number from 0 to 10 meant to express how interesting the finding is. Repute is “good”, “bad” or neutral. Both are assigned by SNPedia editors, not computed from effect sizes, and both are the reason the report reads more confidently than the underlying evidence supports.

The tool does no imputation, no polygenic scoring, and no phasing. It also does no filtering by allele frequency or study quality. It is a join. Promethease has been operated by MyHeritage since 2019, and the report is delivered as a downloadable HTML bundle you can keep offline, which is a genuine advantage over services that hold your results behind a login.

The database is a literature index and not a clinical classification

SNPedia aggregates published associations. Some of those come from ClinVar and large GWAS consortia. Many come from single small candidate-gene studies from the 2000s that were never replicated. Promethease does not distinguish between these tiers in any way you can filter on, so a variant with an odds ratio of 1.06 from a well-powered meta-analysis and a variant from an underpowered study of 200 people can appear with similar magnitudes.

This matters because the interpretive work is where third-party tools add or destroy value. Studies of consumers who ran their raw DNA through third-party interpretation services found that many reported results they found confusing or alarming, and a substantial share brought them to a clinician for follow-up 1. Getting more information from a gene test requires understanding what the test measured and what the reported association predicts 2. Promethease gives you the raw material for that reasoning and none of the scaffolding.

There is a structural reason for this. Direct-to-consumer genetics spans several distinct business models, from research-data brokers to subscription interpretation services, and the economics of each shape what the product is willing to say 3. A $12 report cannot include curation review, clinical confirmation or counseling, and Promethease does not pretend otherwise.

The input file is the real limit

Almost every complaint about Promethease is really a complaint about genotyping arrays. A consumer array measures a few hundred thousand pre-chosen positions out of about 3.1 billion base pairs. It reports nothing about the rest of your genome. It calls single nucleotide positions and performs poorly on insertions, deletions, repeat expansions and copy number variants.

Two specific failure modes should govern how you read any report built on this data.

First, rare variant calls on arrays are unreliable. Array clustering algorithms assign genotypes by grouping intensity signals across a batch of samples. For a variant carried by one person in 10,000, there is no cluster to group into, and heterozygous calls at those probes have a high false positive rate. This is why a clinical laboratory will not accept an array call for a pathogenic variant in a gene like BRCA1 without orthogonal confirmation by sequencing. If Promethease surfaces something that looks clinically serious, treat it as a hypothesis requiring a validated clinical test, and take it to a genetic counselor or physician rather than acting on it.

Second, absence of a finding means nothing. The array did not look at most of the genome. A clean Promethease report is not a clean genome, and the gap between those two statements is where most of the risk of misunderstanding sits 4.

What we would run on the same file instead

If you want more than a join, the following gets you a defensible annotation set in a few hours. We prefer this because every step is inspectable and you can version the reference data.

Convert the raw file to VCF against a matched reference. 23andMe v4 and v5 files are on GRCh37. Use bcftools convert --tsv2vcf raw.txt -f human_g1k_v37.fasta -s YOURNAME -Oz -o you.vcf.gz. Watch two things: 23andMe reports genotypes on the plus strand of the reference but uses -- for no-calls, and it encodes mitochondrial and Y contigs with names that may not match your FASTA. Fix contig names with bcftools annotate --rename-chrs.

Annotate against ClinVar directly. Download the ClinVar VCF from NCBI and run bcftools annotate -a clinvar.vcf.gz -c INFO/CLNSIG,INFO/CLNREVSTAT you.vcf.gz. Then filter to variants with at least two-star review status. That single filter removes most of the noise that makes Promethease reports feel alarming, because it drops single-submitter and conflicting classifications.

Add functional consequence and population frequency with Ensembl VEP: vep -i you.vcf.gz --cache --assembly GRCh37 --af_gnomade --everything --vcf -o annotated.vcf. Frequency is the strongest quick filter you have. A variant labeled pathogenic that appears at 2% allele frequency in gnomAD is almost certainly misclassified or very weakly penetrant.

For common-variant risk, impute and score rather than reading individual SNPs. Upload to the TOPMed or Michigan Imputation Server, which returns dosages for tens of millions of positions with per-variant R² quality metrics. Then pull scoring files from the PGS Catalog and compute with plink2 --score weights.txt cols=+scoresums or with the nf-core pgsc_calc pipeline, which handles strand flips and reference matching for you. Compare your score against an ancestry-matched reference panel, because raw score sums are meaningless without a distribution. Most published scores were derived in European-ancestry cohorts and transfer poorly to other groups, which is a general limitation of applying population genetics to one person 5.

So should you pay the $12?

Yes, if you already have a genotype file and you want a searchable index of what has been published about your specific variants. It is fast, cheap, and the offline HTML output is yours. Read it as a reading list.

No, if you want a picture of your health or expect the report to tell you what to do. The magnitude and repute fields imply a level of judgment the underlying evidence does not support, and consumer genetic reports carry real potential for misinterpretation without professional guidance 4. The framing of consumer genomics has shifted over time from genes as a fixed message to genomes as one data stream among many, and that second framing is the more accurate one 6.

If what you want is coverage rather than annotation of a sparse array, the money is better spent on sequencing. A 30x whole genome gives you the indels, the structural variants and the rare coding variation the array cannot see, and it gives you a file you can re-annotate every year as databases improve.

Questions people also ask

Is Promethease legit? It does what it claims: a genotype-to-SNPedia join, delivered as a report. The service is not deceptive about its method. The criticism it attracts comes from users treating a literature index as a medical result, and from the limits of array data underneath it.

Is Promethease free? No. It costs about $12 per report, paid once, with no subscription. SNPedia itself is free to browse, so you can look up individual rsids by hand at no cost if you only care about a handful.

What does Promethease cost compared to alternatives? Promethease is around $12. Open-source annotation with VEP, ClinVar and plink2 is free but costs you a few hours of setup. Whole-genome sequencing with raw file delivery runs from a few hundred dollars upward and changes what is measurable rather than what is interpreted.

Can Promethease tell me if I have a disease? No. It reports associations from published literature at the positions your array measured. Any finding that looks clinically significant needs confirmation on a validated clinical test and discussion with a clinician or genetic counselor.

Is the report safe to upload elsewhere? The output is a local HTML file, which is the safest arrangement available in this category. Whether you upload the underlying raw genotype file to further third-party services is a separate decision, and worth making deliberately.

Footnotes

  1. Catharine Wang, Tiernan J. Cahill, Andrew Parlato, et al. Consumer use and response to online third‐party raw DNA interpretation services. Molecular Genetics & Genomic Medicine, 2017. https://doi.org/10.1002/mgg3.340 ↩

  2. Erika Check Hayden. How to get the most from a gene test. Nature, 2008. https://doi.org/10.1038/456011a ↩

  3. Scott Thiebes, Philipp A Toussaint, Jaehyeon Ju, et al. Valuable Genomes: Taxonomy and Archetypes of Business Models in Direct-to-Consumer Genetic Testing. Journal of Medical Internet Research, 2020. https://doi.org/10.2196/14890 ↩

  4. Mary A. Majumder, Christi J. Guerrini, Amy L. McGuire. Direct-to-Consumer Genetic Testing: Value and Risk. Annual Review of Medicine, 2021. https://doi.org/10.1146/annurev-med-070119-114727 ↩ ↩2

  5. Elena Fumagalli. Direct-to-Consumer DNA Testing and Product Personalization: One Size Does Not Fit All Genes. 2019. https://doi.org/10.4135/9781526463951 ↩

  6. Paula Saukko. Shifting metaphors in direct-to-consumer genetic testing: from genes as information to genes as big data. New Genetics and Society, 2017. https://doi.org/10.1080/14636778.2017.1354691 ↩

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.