A Better Alternative to Genetic Genie for Analyzing Your Own DNA Data
If you want a Genetic Genie alternative, the best one is a local pipeline you control. Convert your raw array file to VCF with bcftools convert --tsv2vcf, then annotate it with Ensembl VEP against a ClinVar VCF and gnomAD frequencies. Query the result with bcftools view. That takes an afternoon and removes the two problems every upload site shares. You cannot see how they filtered, and you are stuck with whatever genes they chose to report. The deeper limitation is the input file rather than the site. A consumer genotyping array measures a few hundred thousand pre-selected positions, so no amount of better software makes it into a genome.
What Genetic Genie is doing, and what it cannot do
Genetic Genie takes a tab-delimited raw data export holding rsid, chromosome, position, and genotype. It looks up a curated list of SNPs and prints your genotype alongside a short narrative. Its two classic reports are methylation and detox. Both are essentially table joins over a few dozen variants in genes like MTHFR, COMT, and the GST family. GenVue Discovery, its Promethease replacement, is a broader join against annotation databases.
The mechanics are sound. The problem is coverage and calibration. The 23andMe v5 chip is an Illumina Global Screening Array with roughly 650,000 probes, which is about 0.02% of the genome. Positions not on the chip are simply absent, and absence in your file means nothing about your biology. Array genotyping also performs worst exactly where clinical interest is highest, on rare variants, because cluster separation depends on having enough carriers in the calibration set.
Three whole categories are invisible to array-based reports. Small insertions and deletions are largely unrepresented. Copy number and structural variants are not called at all by these pipelines. Repeat expansions are out of reach. Comprehensive sequencing routinely finds causal variants that targeted approaches miss, including structural events and variants outside the positions a panel interrogates.1 If you are trying to explain something specific about yourself, the array file is the wrong instrument.
Rebuilding the report locally
Start by getting your raw file into VCF, because every serious tool speaks VCF. You need the right reference FASTA: 23andMe v4 and v5 exports are on GRCh37 (hg19), and mixing builds silently corrupts every downstream annotation. Check the header of your export before anything else.
# strip comments, normalize MT naming, then convert
grep -v '^#' genome_v5.txt | sed 's/^MT/M/' > geno.tsv
bcftools convert --tsv2vcf geno.tsv \
-c ID,CHROM,POS,AA \
-f hs37d5.fa \
-s ME \
-Oz -o me.grch37.vcf.gz
bcftools index me.grch37.vcf.gz
--tsv2vcf will drop any row whose reported alleles do not match the reference at that coordinate, and it prints the count. Read that number. A high drop rate usually means a build mismatch or a strand convention problem rather than an interesting discovery.
Next, annotate. We prefer VEP over snpEff for this because VEP ships with a --custom mechanism that makes ClinVar and gnomAD joins explicit and auditable.
vep -i me.grch37.vcf.gz --vcf --offline --cache --assembly GRCh37 \
--fork 4 --pick_allele_gene --symbol --canonical --sift b --polyphen b \
--custom clinvar_GRCh37.vcf.gz,ClinVar,vcf,exact,0,CLNSIG,CLNREVSTAT \
--custom gnomad.genomes.r2.1.1.sites.vcf.gz,gnomADg,vcf,exact,0,AF \
--compress_output bgzip -o me.vep.vcf.gz
Then filter. The single most useful filter is ClinVar review status, because a “pathogenic” assertion from one submitter with no evidence criteria is not comparable to a reviewed expert panel classification.
bcftools +split-vep me.vep.vcf.gz -c ClinVar_CLNSIG,ClinVar_CLNREVSTAT,gnomADg_AF:Float -a CSQ \
| bcftools view -i 'ClinVar_CLNSIG~"athogenic" && ClinVar_CLNREVSTAT~"criteria_provided" && gnomADg_AF<0.01' \
| bcftools query -f '%CHROM\t%POS\t%ID\t%REF\t%ALT[\t%GT]\n'
Expect very few rows, and treat each one as a hypothesis rather than a result. Array calls at rare positions have a meaningful false positive rate, so a pathogenic hit here is a candidate for orthogonal confirmation by sequencing. Interpreting any such finding is a conversation for a clinical geneticist or a genetic counselor. A command line is not the place for it.
If you want more positions without new data, run imputation. Upload your VCF to the TOPMed Imputation Server, which phases with Eagle and imputes with Minimac4 against a reference panel of roughly 97,000 samples. Keep only variants with an imputation R² above about 0.8, and remember that imputation reconstructs common haplotypes. It will not recover your rare variants, which is where most of the interesting signal lives.
Where array-based tools mislead in ways that look like results
Some pitfalls are technical. Strand flips at ambiguous A/T and C/G sites will make you homozygous for the wrong allele. No-calls appear as -- in the raw file and become missing genotypes, which some report generators quietly render as reference. Probe design also interacts with variation: a SNP under a probe changes hybridization and can distort the measurement itself, an effect long recognized in sequence-tag and hybridization assays.2
Other pitfalls are interpretive. Reports built on single variants in MTHFR, COMT, or the glutathione S-transferases present common polymorphisms as if each carried a defined consequence. Allele frequencies of 30% to 50% should make you skeptical of any large individual effect. Pharmacogenomics is the sharper example. Clinically actionable CYP2D6 status depends on gene deletions, duplications, and hybrid structures with CYP2D7. None of that is visible from array probes. Tools like PyPGx or Aldy can call star alleles, but they need sequencing reads as input.
Gene models are also a moving target, so the same variant can change consequence annotation between transcript sets and VEP releases. Automated gene prediction and annotation have been revised continuously since the first mammalian genome comparisons.34 Pin your cache version and record it with your outputs.
The alternative that changes the answer and not the interface alone
If your goal is to work with your own molecular data, move to 30x short-read whole-genome sequencing. You get roughly four to five million variants instead of 650,000 probe calls, including indels, and you get reads you can revisit as annotation improves.
A pipeline we would run on a WGS CRAM looks like this. Check coverage uniformity with mosdepth --by 1000 --fast-mode before trusting anything. Call small variants with DeepVariant (--model_type=WGS), which is a straightforward container run and outperforms hand-tuned hard filters on indels. Call structural variants with Manta or dysgu and confirm candidates by eye in IGV, since short-read SV calls carry high false positive rates in repetitive regions. Genotype known repeat expansions with ExpansionHunter. Then annotate with the same VEP command above, using GRCh38 caches.
Long reads are the next step up when a specific region resists resolution. Oxford Nanopore assemblies phase highly heterozygous genomes and reconstruct structural variation that short reads cannot span.5 The capability matters for pharmacogenes, HLA, and segmental duplications. Sequencing-first approaches are also what let researchers separate mechanistically distinct cases that look identical on targeted assays.6
Questions people also ask
Is there a free Genetic Genie alternative?
Yes, and the best one is the open-source stack: bcftools, Ensembl VEP with a local cache, the ClinVar VCF from NCBI, and gnomAD frequency files. All are free for personal use, and everything stays on your machine. The cost is a day of setup and reading documentation.
Can I upload whole-genome sequencing data to these sites?
Mostly no. Genetic Genie, Promethease-style tools, and similar sites expect a small array-format text file. A 30x WGS VCF has several million rows and breaks their assumptions about file size and variant density, so you need your own pipeline anyway.
Should I trust MTHFR or detox reports?
Treat them as genotype lookups rather than assessments. The underlying variants are common, effect estimates from single SNPs are small, and the narrative text on these sites is not a clinical interpretation. If a result concerns you, take it to a clinician who can order confirmatory testing.
Can an AI agent interpret my raw DNA file for me?
An agent is good at the mechanical parts: parsing VCFs, joining annotation tables, checking allele frequencies, and drafting queries. It is not a substitute for clinical classification, and it will confidently over-read array artifacts. Give it reads or a well-filtered VCF and ask it for candidates with supporting evidence.
What replaced Promethease after it shut down public uploads?
GenVue Discovery and codegen.eu fill the same niche with the same input constraints. The durable replacement is a local VEP run, because it survives any one service going offline and lets you re-annotate the same file as databases update.
Footnotes
-
Yumi Enomoto, Takayuki Yokoi, Yoshinori Tsurusaki, et al. Divergent variant patterns among 19 patients with Rubinstein‐Taybi syndrome uncovered by comprehensive genetic analysis including whole genome sequencing. Clinical Genetics, 2022. https://doi.org/10.1111/cge.14103 ↩
-
A. P. M. Silva. The impact of SNPs on the interpretation of SAGE and MPSS experimental data. Nucleic Acids Research, 2004. https://doi.org/10.1093/nar/gkh937 ↩
-
M BRENT. Recent advances in gene structure prediction. Current Opinion in Structural Biology, 2004. https://doi.org/10.1016/s0959-440x(04)00081-8 ↩
-
Mouse Genome Sequencing Consortium. Initial sequencing and comparative analysis of the mouse genome. Nature, 2002. https://doi.org/10.1038/nature01262 ↩
-
R. Jordan Price, Jahn Davik, Felicidad Fernandéz Fernandéz, et al. Chromosome-scale genome sequence assemblies of the ‘Autumn Bliss’ and ‘Malling Jewel’ cultivars of the highly heterozygous red raspberry (Rubus idaeus L.) derived from long-read Oxford Nanopore sequence data. PLOS ONE, 2023. https://doi.org/10.1371/journal.pone.0285756 ↩
-
Carla Stephan, Majd Al Assaad, Max F. Levine, et al. Whole genome sequencing elucidates etiological differences in MCPyV-negative Merkel cell carcinoma. Pathology - Research and Practice, 2024. https://doi.org/10.1016/j.prp.2024.155668 ↩
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.