1 · Upload Genotype Files or Pick a 1000 Genomes Sample
Converted files can be uploaded in the supported text formats. ZIP and GZ are also accepted; the same uncompressed size limit applies.
2 · Scoring Mode & Options
Default: Multi-Trait Genetic-Liability Score & Conditional Risk — weights the direct UKB score and genetically correlated trait scores to predict never-had-sex liability, using published accuracies, genetic correlations and 503 European reference genomes. The predicted gain is model-based; no cohort with genotypes and this outcome was available to test it.
Service usage and registered source-link activity are counted without cookies or individual visitor histories. Landing-page visits also count the referring site's HTTP(S) origin when the browser supplies it; full URLs, paths, queries, fragments and credentials are not retained. These aggregate usage counts measure visits and analyses, not unique people. They are scheduled for deletion after 90 days; cleanup may take longer.
Enabled by default. Uncheck to opt out; your analysis and downloads are unchanged. Your choice is remembered in this browser when storage is available. Sharing saves private individual score records for eligible computed trait models and combined methods, including standardized scores, percentiles, coverage, scoring settings, source and date, linked to a keyed genotype fingerprint. These pseudonymous records are available to the operator for analysis and retained until the operator deletes them. They are separate from the pooled percentile histograms for each model, overall and by source. A later optional survey requires separate consent before any answers are saved.
The private keyed fingerprint recognizes repeat uploads without retaining names, filenames, raw genotypes, IP addresses or full referring URLs. New contributions count each distinct called-genotype dataset once per model across all sources; the first eligible source receives credit for that model. Renaming, compressing or reordering the same calls does not create a new dataset; different chips, subsets or changed calls may. Counts are not guaranteed unique people. Aggregate summaries are released in batches of 20 eligible contributions per model, rounded down to groups of five with small cells withheld. Historical contributions cannot be retroactively deduplicated. Opting out creates no new fingerprint or individual score record, hides the survey, and does not erase data already saved. Bundled public reference demos are excluded.
This site uses hCaptcha, a third-party verification service. Its Privacy Policy and Terms of Service apply.
Elapsed time: 0:00
Keep this tab open. Do not refresh or close it until your results are ready. Large files and additional models may take several minutes.
Side-by-Side Genome Comparison
Public reference data · no upload needed
1000 Genomes Phase 3Average scores across genetic reference groups
Explore mean polygenic scores in five broad reference groups. These are genetic reference labels, not self-reported ethnicities. The map shows score calibration shifts; these values are not validated differences in traits or risk.
Loading public reference-map data…
0 · EUR mean
Mean score in EUR standard deviations (fixed across models and modes)
| Reference group | Mean z | 95% interval | Range across masks | Within-group SD | N |
|---|
All groups use the same 503-genome EUR standardization, so EUR has mean 0 and SD 1 by construction. Within-group SD describes variation among genomes (averaged over masks in chip mode); the 95% interval describes sampling uncertainty in the group mean, not uncertainty in the model or its transfer between groups.
The five regional anchors are symbolic, not sample locations, population boundaries, or user locations. AMR is an admixed reference group. Colors apply only to the markers; no values are interpolated between them.
Source: . Public reference genomes are not a representative sample of all people in these regions.
Why these are score-distribution diagnostics: differences in mean scores depend on the GWAS and scoring methods and cannot be read directly as differences in traits (Duncan et al., 2019).
Study Reference Data & Interpretation Guide (Abdellaoui et al., 2025)
Box 1 · Cautionary Notes on Interpreting Genetic Associations with Sexlessness
- Not “genes for sexlessness”: Polygenic scores aggregate thousands of common variants of minuscule individual effect estimated in a contemporary Western cohort (UK Biobank, ages 39–73). They reflect a culture-specific interplay between heritable traits and social environments, not biological determinism.
- Not predictive for individuals: SNP-based heritability is 17% (males) and 14% (females), but the study's own polygenic score explained only ~0.03% of observed variance in UK Biobank siblings (≈0.3% on the liability scale; Dataset S1 Tab 5). The authors state that these scores "are not suited to predict individual-level outcomes regarding sexlessness".
- Ancestry calibration: Weights come from European-ancestry cohorts and percentiles are relative to the 503 European genomes of 1000 Genomes. For other ancestries they shift systematically (often by more than 1 SD; see the calibration table below) and should not be interpreted.
- Socioecological context matters: By current address, male sexlessness was higher in regions with relatively fewer women (r = −0.07, P = 0.0004), and sexlessness was higher in regions with greater income inequality (Gini; pooled r = 0.15, P = 9.5×10−13; males r = 0.09, females r = 0.14). Associations by birthplace were not significant.
Genetic Correlations (rg) with Sexlessness Across Complex Traits (Fig. 3 & Dataset S1 Tab 7)
Selected LDSC genetic correlations between sexlessness (pooled, males, females) and 82 traits from Dataset S1 Tab 7. ‡ In the source table the reported rg and z have opposite signs for these traits (coding inconsistency); values are shown as published.
Combined Model & Risk Calibration Evidence
The original PGS002215 “no_sex” model had a covariate-adjusted partial correlation of 0.0313 (95% CI 0.0166–0.0460) in 17,730 held-out UK/Ireland participants (PGS Catalog, PPM012567). Its original training phenotype contained 3,320 cases among 347,296 participants. These data inform the risk scenarios; they do not validate the combined model. The displayed sensitivity range propagates the validation correlation bounds under fixed assumptions, not all sources of individual prediction uncertainty.
The default multi-trait score predicts never-had-sex liability with weights proportional to S−1c, where S is the reference covariance of the chip scores and ck = (score accuracy for its own trait) × (published genetic correlation with sexlessness) × (target SNP-heritability root). Accuracies come from published held-out evaluations or from summary-statistic associations with the Mills 2021 AFS/AFB GWAS. That identity reproduced observed associations implied by published AFS/AFB genetic correlations (slope 0.95, r = 0.88 across 14 checks). A Mills AFB summary-statistic probe and reference-genome covariance both place the direct score's accuracy below its published held-out anchor; the model combines all three (equivalent partial r ≈ 0.025 versus 0.0313 published). The model-based liability R² is several times that of the direct score alone, but this gain has not been measured in an outcome cohort. Build and validation details: reference/outcome_prediction/multitrait/validation.json.
The legacy trait-prior index uses published genetic correlations as heuristic priorities. Within-reference score correlations are not converted into genetic correlations. Related trait scores without validated genetic correlations remain available individually.
Out-of-Sample Polygenic Score Validation in the Australian Twin Cohort (Table 2 & Dataset S1 Tab 6)
Generalized estimating equation (GEE) results for the study's own sexlessness PGS in an independent Australian cohort (QIMR Berghofer, ages 18–89, N = 1,354–13,532). This table does not validate PGS002215 or this app's combined score.
1000 Genomes Phase 3 Reference Panel Calibration (N = 2,504 Real Genomes · PLINK 2 Verified)
z-score mean ± SD per super-population. Scores are standardised empirically against 1000 Genomes EUR on the exact SNPs present; the EUR column is out of sample (standardised on a random half, evaluated on the other half). "LD var. ratio" = empirical EUR variance ÷ the independence formula Σ2p(1−p)β²: values far above 1 show why finished PGS cannot be standardised analytically. Raw sums match PLINK 2 --score.