Interpretation reference

Yeliztli's modules produce findings in a handful of common "languages." This page explains the evidence rating that accompanies every finding, the different kinds of output you'll see, and the recurring caveats that apply across modules. It complements the per-module pages and the gentler reading your results overview.

Not a medical test — for research and educational use only

Yeliztli analyses consumer genotyping-array data (23andMe / AncestryDNA), which is not a clinical-grade test. Results are not diagnostic, are not clinically validated, and must not be used to make medical decisions. Array data is especially unreliable for rare, disease-causing variants. Always confirm any finding with an accredited clinical laboratory and discuss it with a qualified clinician or genetic counsellor before acting on it. See the Intended use & disclaimers page for the details and the evidence behind this warning.

Evidence ratings (★ to ★★★★)

Every finding carries a star rating so you can weigh it. The bands map to established evidence frameworks:

Rating	Roughly corresponds to
★★★★	ClinVar Pathogenic / Likely-Pathogenic with a reviewed (2+ star) status, CPIC Level A, or a genome-wide-significant GWAS association with a very large effect size (for example, odds ratio > 5)
★★★	ClinVar Pathogenic / Likely-Pathogenic (single submitter), CPIC Level B, or a replicated, genome-wide-significant GWAS association
★★	A variant of uncertain significance with functional support, a single genome-wide-significant GWAS association without independent replication, or PharmGKB level 2A/2B
★	A single study, a candidate-gene association, PharmGKB level 3/4, or a carried rare/novel variant surfaced for discovery context without stronger ClinVar or functional evidence

Not every ★ row is a disease association. The Rare Variant Finder also uses ★ for carried rare, AF-missing, or novel variants that pass its discovery filters but lack stronger clinical or functional evidence. Treat those rows as a low-evidence variant inventory until you review the variant details and any linked evidence.

For GWAS findings, the conventional p < 5×10⁻⁸ threshold controls genome-wide multiple testing, but it does not by itself make an association definitive. Yeliztli reserves higher GWAS tiers for genome-wide significance plus independent replication, or for genome-wide significance plus a very large effect size under Yeliztli's tiering rule. The cited GWAS papers support the p-value, replication, and false-positive-control rationale [1,2].

Two rules keep weak signals from looking strong:

• Wellness/trait modules are capped (often at ★★). The Traits & personality module, for example, caps every finding regardless of the underlying variant.
• Weak variants can't escalate a category. In categorical modules, a ★ variant cannot push a pathway to Elevated (it's held at Moderate).

Kinds of output

Pathogenic-variant findings

The hereditary-risk modules (cancer, cardiovascular, carrier status) report specific variants classified by ClinVar using the ACMG/AMP framework. The Rare Variant Finder overlaps with that when a carried rare variant has a ClinVar Pathogenic/Likely-Pathogenic assertion, but it also reports a broader inventory of carried rare, AF-missing, ensemble-pathogenic, and novel variants. Only its ClinVar pathogenic categories should be read as pathogenic-variant findings. Read these with the inheritance pattern in mind:

• Carrier vs affected. For recessive conditions, carrying one copy makes you a carrier (usually unaffected); two copies is an affected state. Yeliztli labels which applies.
• Cross-links. Some genes appear in more than one module (e.g. BRCA1/2 in both cancer and carrier status) with framing appropriate to each.
• Negatives aren't clearance. Arrays type specific variants, not whole genes — a negative result doesn't exclude untyped variants.

ClinVar classifications that conflict

Yeliztli reports pathogenic-variant findings only when ClinVar has a clear Pathogenic/Likely-Pathogenic classification. It deliberately does not turn records labelled Conflicting classifications of pathogenicity into findings, because the submitted clinical interpretations disagree and should not be presented as a definitive finding.⁵

That means "no findings" does not mean "no variants with any pathogenic-leaning ClinVar evidence." To review contested records, open the Variant Explorer and filter or search ClinVar significance for conflicting.

In-silico pathogenicity scores (CADD, REVEL)

Variant surfaces — the Variant Explorer side panel and the Rare Variant Finder — show two computational pathogenicity predictors as raw numbers. Unlike SIFT and PolyPhen-2 (shown with a plain-language Deleterious / Tolerated label), CADD and REVEL are unlabelled, so their direction and scale are:

• CADD (Combined Annotation-Dependent Depletion) — a phred-scaled measure of variant deleteriousness: higher = more deleterious, on a scale of roughly 0–99. A CADD-Phred of 10, 20, or 30 corresponds to the ~top 10%, 1%, or 0.1% most deleterious of all possible single-nucleotide variants.⁸
• REVEL (Rare Exome Variant Ensemble Learner) — a missense-specific ensemble score on a 0–1 scale: higher = more likely pathogenic.⁹

The Rare Variant Finder highlights a score in red above the thresholds it uses for display — CADD ≥ 20 and REVEL ≥ 0.5. These are display heuristics to draw attention, not diagnostic cut-offs: in-silico predictions are supporting evidence only, to be weighed alongside ClinVar, allele frequency, and inheritance — never read as a diagnosis on their own.

Categorical pathway levels

The wellness modules (nutrigenomics, methylation, fitness, sleep, skin, allergy, traits) and gene health report a level per pathway — Elevated, Moderate, or Standard — rather than a number. A pathway's level reflects its highest-category contributing variant, subject to the evidence cap above. Sites that can't be resolved from array data (e.g. strand-ambiguous palindromic homozygotes) are marked Indeterminate and withheld from the level rather than guessed.

When your array only partially covers a pathway's tracked SNPs — common across genotyping vendors — a Standard pathway carries a coverage-qualified badge so the call isn't read as fully supported:

• Tested Standard — the pathway looks Standard, but only some of its tracked SNPs were on-chip or callable from your array; the rest were not assessed. It is a Standard call made on incomplete coverage (the card's coverage note says how many were off-chip vs. no-call).
• Not Assessed — none of the pathway's tracked SNPs were callable, so no level could be determined for that pathway.

These differ from a plain Standard badge (all tracked SNPs covered) and from Indeterminate (individual sites that were typed but can't be resolved, withheld from an otherwise-computed level). Elevated and Moderate pathways always show their plain level, never a coverage-qualified badge.

Star-allele diplotypes & CPIC status

Pharmacogenomics reports a diplotype (e.g. *1/*4) and a CPIC phenotype or functional status, each with a call-confidence of Complete, Partial, or Insufficient. Genes reported with metabolizer phenotypes use Poor Metabolizer, Intermediate Metabolizer, Normal Metabolizer, Rapid Metabolizer, and Ultrarapid Metabolizer. Rapid is a distinct increased-activity category, not a synonym for ultrarapid; for example, Yeliztli maps CYP2C19 *1/*17 to Rapid Metabolizer and *17/*17 to Ultrarapid Metabolizer.⁶ Other CPIC pharmacogenes use different status families, such as Normal function, Decreased function, and Poor function for SLCO1B1, or Rapid Acetylator, Intermediate Acetylator, and Slow Acetylator for NAT2. Some PGx results also include an activity score. Treat it as a gene-specific support value for the phenotype/status, not as a universal scale: activity-score thresholds are defined per gene and diplotype system, and the same number can map to different interpretations in different pharmacogenes.⁷ Drug alerts use CPIC levels. Treat Partial/Insufficient calls with extra caution — arrays can miss copy-number and structural variation.

Polygenic scores (percentiles)

The polygenic modules (metabolic, bone density, familial hypercholesterolemia, cancer PRS, cognitive traits) summarise many small-effect variants into a population percentile, never a raw score or a PRS-derived absolute risk. Read these carefully:

• Research use only. They are not diagnostic and not clinically validated.
• Percentiles can be withheld. When a score isn't calibrated for your inferred ancestry, or your array covers too few of its variants, Yeliztli withholds the percentile and shows coverage instead — rather than report a misleading number.
• Ancestry matters — a lot. Scores are mostly derived in European-ancestry cohorts and transfer poorly to other ancestries, which can worsen health disparities if misused ¹. Accuracy in fact decays continuously with genetic distance from the training data — even within a single labelled ancestry group ² — and varies with non-genetic factors such as age, sex, and socio-economic status ³⁴. Yeliztli shows an ancestry-mismatch warning when relevant.

This PRS rule is separate from the cancer module's opt-in absolute-risk context, which can show population baseline and monogenic BRCA1/BRCA2 carrier-penetrance figures after explicit consent. That overlay is not a PRS-derived personal risk estimate.

Risk-genotype findings (common variants)

Several condition modules (e.g. haemochromatosis, thrombophilia, AMD, gout, APOL1) report common risk genotypes with odds ratios. These describe relative risk; the absolute lifetime risk for most carriers often remains low, penetrance is frequently reduced, and some effects are ancestry- or sex-stratified. They are risk modifiers, not diagnoses.

Haplogroups & ancestry

Ancestry estimates, admixture fractions, and haplogroups are statistical inferences — see Ancestry methods & validation for how they're computed and their limitations.

Recurring caveats

• Array-proxy genotyping. Some variants are read via a nearby proxy SNP, not directly; proxy accuracy varies by ancestry. And array calls for rare variants are unreliable in general — see Intended use & disclaimers.
• Reduced penetrance. Carrying a risk genotype is not the same as having (or developing) the condition.
• Sex/ancestry stratification. Where penetrance differs by sex or ancestry, Yeliztli says so.
• Honest gaps. When a result can't be supported (an uncalibrated percentile, an unresolvable palindromic call), Yeliztli withholds it instead of fabricating one.

Module → output type

Output type	Modules
Pathogenic variants (ClinVar P/LP)	Cancer, Cardiovascular, Carrier status, Rare variants (ClinVar pathogenic category)
ClinVar lower-penetrance/risk-allele variants	Rare variants
Carried rare/novel variant inventory (mostly ★ discovery context)	Rare variants
Categorical pathway levels	Nutrigenomics, Methylation, Fitness, Sleep, Skin, Allergy, Gene health
Star-allele diplotype + CPIC phenotype/status	Pharmacogenomics
Polygenic percentile	Metabolic, Bone density, Familial hypercholesterolemia, Cancer (PRS), Traits
Common-variant odds ratios	Haemochromatosis, Thrombophilia, Alpha-1, AMD, APOL1, Gout
Diplotype / risk genotype (gated)	APOE, Parkinson's, Sex-aneuploidy
Mitochondrial / pharmacogenomic risk	LHON, MT-RNR1, G6PD, BChE
Ancestry & haplogroups	Ancestry

References

[1] Barsh GS, Copenhaver GP, Gibson G, Williams SM. Guidelines for Genome-Wide Association Studies. PLOS Genetics. 2012;8(7):e1002812.

[2] Chen Z, Boehnke M, Wen X, Mukherjee B. Revisiting the genome-wide significance threshold for common variant GWAS. G3: Genes, Genomes, Genetics. 2021;11(2):jkaa056.

---

1. Principles and methods for transferring polygenic risk scores across global populations (Kachuri et al., 2023, Nature Reviews Genetics). ↩

2. Polygenic scoring accuracy varies across the genetic ancestry continuum (Ding et al., 2023, Nature). ↩

3. Variable prediction accuracy of polygenic scores within an ancestry group (Mostafavi et al., 2019, eLife). ↩

4. Portability of 245 polygenic scores when derived from the UK Biobank and applied to 9 ancestry groups from the same cohort (Privé et al., 2022, Am. J. Hum. Genet.). ↩

5. ClinVar: improving access to variant interpretations and supporting evidence (Landrum et al., 2018, Nucleic Acids Research) describes ClinVar as a public archive of submitted clinical-significance interpretations for human variants. ↩

6. Standardizing terms for clinical pharmacogenetic test results: consensus terms from the Clinical Pharmacogenetics Implementation Consortium (CPIC) (Caudle et al., 2017, Genetics in Medicine) defines consensus pharmacogenetic phenotype terminology for consistent PGx interpretation. ↩

7. Standardizing CYP2D6 Genotype to Phenotype Translation: Consensus Recommendations from the Clinical Pharmacogenetics Implementation Consortium and Dutch Pharmacogenetics Working Group (Caudle et al., 2020, Clinical and Translational Science; PMID 31647186; PMCID PMC6951851) describes the CYP2D6 activity-score system as summed allele activity values and explains that phenotype translation depends on consensus, gene-specific thresholds. ↩

8. CADD: predicting the deleteriousness of variants throughout the human genome (Rentzsch et al., 2019, Nucleic Acids Research; PMID 30371827) describes CADD as a phred-scaled, genome-wide measure of variant deleteriousness in which higher scores indicate more deleterious variants. ↩

9. REVEL: An Ensemble Method for Predicting the Pathogenicity of Rare Missense Variants (Ioannidis et al., 2016, American Journal of Human Genetics; PMID 27666373) presents REVEL as a missense-specific ensemble score from 0 to 1 in which higher scores indicate a greater probability of pathogenicity. ↩