Free raw data interpretation: what the upload sites can and cannot tell you
The best free interpretation of a consumer DNA file is the one you run yourself: convert the file to VCF with bcftools convert --tsv2vcf, annotate it with Ensembl VEP against a local cache, and cross-reference ClinVar and gnomAD. That takes an evening and gives you a reproducible pipeline instead of a PDF. The upload sites Promethease, Genetic Genie, Sequencing.com and Codegen all do roughly the same thing with less transparency. Every one of them inherits the same ceiling. A genotyping array measures a few hundred thousand pre-chosen positions, so most of what you might want to know is simply not in the file.
That ceiling matters more than the choice of tool. Below we walk through the pipeline we would run, then the specific ways array data misleads people, then what changes when the input is sequencing rather than genotyping.
What the free upload sites do
Promethease was the reference implementation for years. It matched rsIDs in your file against SNPedia and returned a ranked list of “genotypes” with literature blurbs. It is now part of MyHeritage and charges a small fee. Genetic Genie produces methylation and detox reports built on a handful of SNPs. Sequencing.com wraps uploads in an app store of third-party reports.
All of them perform lookup and presentation. They do not perform variant calling. Your file already contains genotype calls made by the vendor’s clustering algorithm. The report layer adds annotation text. Third-party interpretation services vary widely in analytic quality and in how they communicate uncertainty, and reviewers have flagged the ethical problem of returning health-adjacent claims without clinical context 1. Treat any output from them as a hypothesis list. It is not a result.
The pipeline we would run on a 23andMe or AncestryDNA file
A 23andMe export is a tab-separated file with a comment header and four columns: rsid, chromosome, position and genotype. Positions are on GRCh37 for v3 through v5 chips. AncestryDNA splits the genotype into two allele columns, so it needs one awk pass first.
Convert to VCF with a reference FASTA so that reference and alternate alleles are assigned correctly:
bcftools convert --tsv2vcf genome_v5.txt \
-f human_g1k_v37.fasta -s ME -Oz -o me.b37.vcf.gz
bcftools index me.b37.vcf.gz
The default column mapping (-c ID,CHROM,POS,AA) matches the 23andMe layout. The step is doing real work: it resolves each genotype against the reference base, drops sites where the alleles are inconsistent, and turns -- no-calls into missing data rather than silent reference calls.
Most current annotation resources are on GRCh38, so lift over before annotating:
CrossMap.py vcf hg19ToHg38.over.chain.gz me.b37.vcf.gz \
GRCh38.fa me.b38.vcf
Then annotate offline with VEP and a local ClinVar VCF:
vep -i me.b38.vcf -o me.annot.vcf --vcf --offline --cache \
--assembly GRCh38 --everything --pick_allele_gene \
--custom clinvar.vcf.gz,ClinVar,vcf,exact,0,CLNSIG,CLNREVSTAT
Finally, filter to the small set of sites worth a second look:
bcftools view -i 'INFO/CSQ~"Pathogenic" && GT!="0/0"' me.annot.vcf
Running this offline keeps your genotypes on your own disk and makes the analysis reproducible, which is the same argument made for pipeline-based genomics generally 2. Surveys of variant analysis tooling are a good map if you want alternatives to VEP, such as ANNOVAR or SnpEff 3.
Expect the filtered output to be short and mostly disappointing. That is the correct result.
Where array data fails
Four failure modes account for nearly every confusing result people bring to forums.
No-calls. Roughly one to two percent of probes fail on any given chip run. A -- genotype is missing data. It is not homozygous reference. Any tool that treats absence as normal will quietly reassure you about a site it never measured.
Strand ambiguity. Illumina arrays report alleles on the chip’s own strand convention, which is not always the same as the reference forward strand. For A/T and C/G sites the flip is undetectable from the genotype alone. This is the main reason a variant appears “reversed” compared to a published paper. The FASTA-based conversion above catches most cases but cannot resolve palindromic sites.
Indels. 23andMe encodes insertions and deletions as I and D rather than sequence. These do not convert cleanly and are usually best dropped.
Rare-variant false positives. Arrays cluster fluorescence intensities into three genotype classes. Clustering works well when the minor allele is common and badly when it is very rare, because there are almost no true heterozygotes in the training data to anchor the cluster. The practical consequence is that a rare pathogenic call from a consumer array has a low positive predictive value. If a filtered hit matters to you, it needs orthogonal confirmation by a clinical laboratory, ordered through a clinician who can interpret it. Clinical reporting standards exist precisely because analytic validity and evidence review are separate problems 4.
What changes with sequencing
The array measures sites someone chose in advance. Whole-genome sequencing measures the genome and lets you choose afterward, which is the difference between a fixed panel and a reusable dataset 5. Short-read sequencing at around 30x coverage yields four to five million variants per person, including the rare and novel ones arrays cannot represent 6.
Sequencing also gives you a calibration path. Genome in a Bottle reference materials provide high-confidence variant call sets for benchmark samples, so a pipeline’s sensitivity and precision can be measured rather than assumed 7. Tools like SpeedSeq showed years ago that alignment through calling and annotation can run end to end on commodity hardware in about a day 8. If you keep the aligned CRAM rather than only a report, the same data supports reanalysis as ClinVar and gnomAD improve, which is where most of the long-term value sits 9.
MTHFR and similar single-SNP reports
The two variants behind most free methylation reports are rs1801133 (C677T) and rs1801131 (A1298C) in MTHFR. Both are common. Your file contains them, and reading them out is trivial:
bcftools view -i 'ID=="rs1801133" || ID=="rs1801131"' me.b38.vcf.gz
What the genotype means functionally is a separate question, and the reports that assign supplement protocols to these genotypes are extrapolating well past what the genotype supports. Functional genomics generally requires expression, protein, and metabolite measurements to say anything about an individual pathway 10. If a homocysteine or folate question is on your mind, that is a blood measurement and a conversation with a clinician. A SNP lookup will not answer it.
Privacy before you upload
Every free interpretation site is a data transfer. Before uploading, check whether the service stores your file and whether deletion is real. Also check what happens to the data if the company is sold. The 2023 credential-stuffing incident at 23andMe and the company’s later bankruptcy proceedings made the ownership question concrete for millions of people. Running VEP locally avoids the question entirely, which is our main reason for recommending it.
Questions people also ask
What is the best free tool for raw DNA analysis? Ensembl VEP with a local cache, driven by the bcftools conversion above. It is free, offline and versioned, and it is used in clinical pipelines. If you want a browser tool instead, expect a lookup table rather than analysis.
What is a free alternative to Promethease? The self-hosted pipeline described here. For a hosted option, Genetic Genie and Sequencing.com offer free tiers, with the caveats about interpretation quality noted above 1.
Is GEDmatch really free? The core matching and admixture tools are free. Tier 1 features sit behind a monthly subscription. GEDmatch is a relative-matching service and does not do health interpretation.
What is the Ancestry.com scandal? There is no single event. The phrase usually refers to a cluster of privacy disputes across consumer genetics: broad terms of service around research use, law enforcement access to genealogy databases, and changes of corporate ownership. Read the current terms before uploading anywhere.
Which DNA site is most accurate? For common variants, the major consumer arrays agree closely with each other and with sequencing. For rare variants, none of them are accurate enough to act on without confirmation in a clinical laboratory.
Footnotes
-
Lauren Badalato, Louiza Kalokairinou, Pascal Borry. Third party interpretation of raw genetic data: an ethical exploration. European Journal of Human Genetics, 2017. https://doi.org/10.1038/ejhg.2017.126 ↩ ↩2
-
Anton Nekrutenko, James Taylor. Next-generation sequencing data interpretation: enhancing reproducibility and accessibility. Nature Reviews Genetics, 2012. https://doi.org/10.1038/nrg3305 ↩
-
S. Pabinger, A. Dander, M. Fischer, et al. A survey of tools for variant analysis of next-generation genome sequencing data. Briefings in Bioinformatics, 2013. https://doi.org/10.1093/bib/bbs086 ↩
-
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 ↩
-
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 ↩
-
Jason M. Rizzo, Michael J. Buck. Key Principles and Clinical Applications of “Next-Generation” DNA Sequencing. Cancer Prevention Research, 2012. https://doi.org/10.1158/1940-6207.capr-11-0432 ↩
-
Justin M. Zook, David Catoe, Jennifer McDaniel, et al. Extensive sequencing of seven human genomes to characterize benchmark reference materials. 2015. https://doi.org/10.1101/026468 ↩
-
Colby Chiang, Ryan M Layer, Gregory G Faust, et al. SpeedSeq: ultra-fast personal genome analysis and interpretation. Nature Methods, 2015. https://doi.org/10.1038/nmeth.3505 ↩
-
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 ↩
-
Evelina Gasperskaja, Vaidutis Kučinskas. The most common technologies and tools for functional genome analysis. Acta medica Lituanica, 2017. https://doi.org/10.6001/actamedica.v24i1.3457 ↩
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.