How Pathway thinks.
Every recommendation Pathway makes traces back to a specific variant in your file, a specific rule in an open scoring engine, and a conservative dose range — not a fixed dose. This page walks through how that pipeline works, what it deliberately doesn't do, and why.
What Pathway analyzes
32 single-nucleotide variants across 7 biological pathways. A small, conservative list on purpose. Third-party tools that report on hundreds of variants tend to score noise; variants with genuinely actionable literature are a much smaller set.
Methylation
8 variants · MTHFR, MTRR, MTR, CBS, BHMT, MTHFD1, PEMTHow your body activates folate and B12, recycles homocysteine, and produces the phospholipids that keep membranes and neurotransmitters working. The best-studied pathway in consumer genetics; also the one most prone to over-interpretation.
Histamine
2 variants · HNMT, CYP1A2How efficiently your body breaks down histamine and caffeine — relevant to sleep, migraines, and food-histamine reactions in a subset of people.
Neuro
4 variants · COMT, BDNFHow you clear dopamine and adrenaline, and how quickly you build new synaptic connections. Small effect sizes individually; interesting when clustered.
Vitamin D
4 variants · CYP2R1, VDR, GCHow you convert sunlight into circulating vitamin D and how efficiently your cells use it. A common reason two people with the same intake end up at different blood levels.
Inflammation
5 variants · TNF, IL6, TGFB1, IL10Baseline set-points on your innate-immune signaling. Not a disease diagnosis — a tilt toward how strongly your body tends to inflame and how quickly it resolves.
Blood sugar & metabolic
4 variants · TCF7L2, MTNR1BTwo of the most-replicated common variants tied to type-2-diabetes risk and evening glucose control. Score more usefully together than alone.
Pharmacogenomics
5 variants · CYP2C19, SLCO1B1, OPRM1, ABCB1, CYP1B1How you metabolize common drug classes — SSRIs, statins, opioids, and P-glycoprotein substrates. Surfaces as flags for your doctor to consider, not as recommendations to change medication.
The full variant-by-variant list — with gene, variant name, and a link to the canonical dbSNP entry for each — lives on the citations page.
How scoring works
1 · Read your file locally
Your raw 23andMe / AncestryDNA archive is parsed in your browser. The scoring engine only ever sees the 32 variant calls it needs. Nothing else from the file is retained — see the security page.
2 · Normalize strand orientation
Different providers report the same variant on different strands (23andMe reports several MTHFR calls on the reverse strand, for example). The parser flips each call to a canonical forward-strand form before any scoring runs, so “CT” means the same thing regardless of which kit you sent in.
3 · Classify each call
Each variant lands in one of three buckets: clear (two reference copies — no adjustment needed), heterozygous (one variant copy — partial impact), or homozygous risk (two variant copies — full impact). Buckets are pre-declared per variant, not inferred at runtime.
4 · Assign a tier
Each finding is scored as either essential (strong literature, meaningful effect size, worth prioritizing) or optimal (real but smaller effect, or emerging evidence). A few variants are pharmacogenomic flags — surfaced as “discuss with your doctor” rather than supplement recommendations. Clear calls are never scored.
5 · Apply pathway rules
Some findings only become meaningful in combination. A compound heterozygote for both common MTHFR variants (rs1801133 + rs1801131) gets an essential-tier methylation finding even though each variant alone is only optimal-tier. These rules are named and cited in the scoring engine, not hidden.
6 · Turn findings into a stack
Multiple findings often point to the same supplement. Pathway deduplicates so you get one methylfolate recommendation, not three, with all the supporting pathways cited on it. Every item in the stack has a reason button that walks back to the variant and rule that put it there.
Conservative defaults
Where the literature is ambiguous, Pathway defaults to the safer interpretation. Some examples baked into the engine:
- Dose ranges, not fixed doses. Every recommendation is a range (e.g. “400–800 mcg”) so you and your clinician can start at the lower end and adjust.
- Start-low protocols. For variants where an over-methylation reaction is plausible (methylfolate + slow COMT), the recommendation explicitly says to start at the low end and titrate.
- Homozygous-only for CBS. Heterozygous CBS calls are common and not clinically meaningful in most people; only the homozygous-risk genotype is scored, and only as an optimal-tier finding.
- Verified vs. assumed strand orientation. Each variant carries an internal flag for whether its provider-strand orientation has been directly verified against a real file or is inferred from dbSNP. Assumptions get flagged in the engine so they can be validated before findings ship.
- Never a diagnosis. No finding is worded as a medical condition; findings describe enzymatic tendencies and cofactor needs.
What Pathway will not do
- Diagnose, treat, cure, or prescribe. This is educational software.
- Apply a one-size-fits-all supplement protocol. Every recommendation traces back to your specific variant calls.
- Report on the “popular” polygenic scores that don't have consistent replication (personality, IQ, athletic ability, dozens of blood-type-diet claims). Not because they're controversial; because the literature isn't strong enough.
- Sell or share your DNA data.