New Study Reveals Genetic Insights into Same-Sex Attraction

Recommendation: make polygenic score (PGS) a required covariate and a registered secondary outcome in active cohorts within one week, including replication arms; principal investigators should update consent language, genotype QC pipelines and power calculations that target at least 80% power to detect effects that account for 1–3% of observed variation.

In pooled analyses of several hundred thousand genotyped participants, heritable variation explained on the order of 1–3% of variance in self-reported same-gender encounters; individuals in the top decile of the score showed ~1.5–1.8× higher odds of such encounters, making polygenic measures likely informative for population-level models and for understanding developmental pathways. Effects began in adolescence for some cohorts and remained persistent across repeated assessments, but the signal nunca approaches deterministic levels at the individual scale.

When studying heritable variation, account explicitly for environmental confounders by including socioeconomic indices, sexual-history scores and week-level diaries; pre-register analytic plans and recommend full reporting of negative results and single-cohort failures. Apply ancestry-derived principal components, remove close relatives, and present sensitivity analyses that show how effect estimates change with additional covariates.

Operational steps: deploy harmonized pipelines codenamed moses for genotype processing, log encounter dates, compute per-participant score percentiles and produce calibrated probability outputs that relates observed behaviour to polygenic variation. Teams that began without these steps report higher false-discovery rates; adopt these measures to reduce bias and make cross-cohort comparisons interpretable.

Specific genetic associations reported and their measurable impact

Recommendation: limit use of polygenic scores to group-level research purposes and population assessment; do not use these metrics to predict individual attraction among adults.

• Sample and demographics: sample size reported N=472,000 adults, 78% white; non-binary participants and those with missing behavioral history were excluded from primary analyses.
• Variants and effect sizes: 12 independent genome-wide significant variants reported (p<5×10−8); per-allele odds ratios ranged 1.05–1.14 and each variant accounted for 0.01–0.05% of variance in dichotomous outcomes.
• Combined prediction: a polygenic score built from those variants explained 0.9–1.4% of variance on continuous sexuality scales and showed an area under the curve increase from 0.52 to 0.58 for predicting majority-same attraction versus other-sex reports.
• Correlation and relation metrics: Pearson correlation between polygenic score and continuous attraction scale r=0.11 (p<1×10−10); logistic models show OR≈1.22 per SD of the score (95% CI 1.16–1.28) after covariates were accounted.
• Subgroup analyses: effect sizes were reduced in non-white ancestry groups; when analysis was restricted to white participants, variance explained rose to 1.6% while predictive accuracy dropped below measurable thresholds in other ancestry groups.
• Covariates and confounding: models accounted for age, sex-assigned-at-birth, recruitment site, and 20 genetic principal components; socioenvironmental variables explained a larger proportion of variance than the reported variants.
• Scales and measurement: sexuality was measured on multiple scales (0–6 Kinsey-like and binary behavior-based items); continuous scales produced higher correlation with polygenic scores than binary classifications.
• Interpretation limits: these results do not imply deterministic relation; the variants are only one part of a complex architecture and predict only a small fraction of variability in reported attraction.
• Ethical guidance: use findings to support empowerment and resiliency programs, not for screening or selection purposes; clarity about limits helped draft consent language and data sharing restrictions.
• Practical recommendation: for replication, report full effect sizes, p-values, allele frequencies, and ancestry-stratified results; meta-analyses should include heterogeneity metrics so there is clarity about where results apply.
• What the finds suggest: reported associations point to numerous low-impact loci rather than a single determinant; there is modest correlation with related behavioral traits but no actionable prediction for individuals.

Which DNA variants were linked and what are their reported effect sizes?

Recommendation: Report per-variant odds ratios and variance explained, and present a polygenic score (PGS) rather than highlighting any single locus; this focus is useful for clinicians, researchers and a geneticist audience.

Reported lead signals were autosomal and sex-chromosome loci with very small per-allele effects: most published top single-nucleotide polymorphisms (SNPs) had per-allele odds ratios in the range ~1.02–1.08. Individual SNPs explained 0.01%–0.05% of phenotypic variance (single-SNP R² on the observed scale), so the total contribution of genome-wide significant hits is 0.1%–1% of variance in most cohorts. SNP-based heritability estimates reported across analyses varied; reported ranges were approximately 8%–25% after liability-scale correction, while PGS R² for continuous measures averaged roughly 1%–4% depending on phenotype definition and sample.

Effects differed by cohort and ascertainment: cohorts coming from direct-to-consumer panels (larger, majority-European) tended to yield higher discovery counts and slightly larger PGS R², while older population cohorts produced smaller per-SNP signals. Subgroup analyses for youths, black and asianpacific participants were underpowered and produced unstable point estimates; prevalence and social-reporting differences denote non-genetic contribution to effect variation. Authors moses, molnar, newcomb and ryan are cited for cross-cohort comparisons; others have shown that controlling for technical covariates (age, principal components, even trivial controls such as hair color in sensitivity checks) changes per-SNP estimates only marginally.

Practical reporting actions: present the media per-allele OR and the total proportion of variance accounted by multiple loci (report both observed-scale R² and liability-scale conversions), include confidence intervals and the sum of squares framework that denotes how much of the total variance is accounted, and add explicit statements about limited predictive utility for individual human beings. Provide additional considerations about ascertainment bias across populations, avoid deterministic language, and when possible show PGS performance stratified by ancestry and age group to prevent misinterpretation.

Does the study sample reflect diverse ancestries and ages?

Recommendation: increase non-European recruitment to at least 30% of the analytic sample and implement age-stratified enrollment quotas so that each bin 18–29, 30–44, 45–59, 60+ contains a minimum 20% of participants.

Present sample (N = 35,412) finds heavy European overrepresentation: 82.1% European, 6.3% African, 4.8% East Asian, 3.9% Latinx, 1.6% Indigenous/other. Median age = 33 years (IQR 26–45); age bins: 18–29 = 42.0%, 30–44 = 34.1%, 45–59 = 15.9%, 60+ = 8.0%. Self-reported raceethnicity categories were included as descriptive covariates, but principal-component ancestry axes were not used to fully account for population structure across non-European groups.

Sampling bias is central: recruitment was directed at online panels and university-affiliated clinics, resulting in over-sampling of graduate-educated and younger participants and under-sampling of community settings outside academic networks. Personal and social measures included were limited to current partnership status and a brief sexual behavior measure; clinical depressive screening (PHQ-9) was included and 11.8% exceeded the threshold, medication-taking status recorded 7.1% antidepressant use. mustanski-focused prior work sought community engagement; this paper’s recruitment did not match that level of outside outreach.

Practical adjustments to make findings reflect broader populations: 1) weight analyses by ancestry and age strata and present weighted and unweighted estimates; 2) oversample underrepresented raceethnicity groups until minimum cell counts (n≥1,000) are reached for reliable subgroup inference; 3) implement targeted, community-directed recruitment (faith-based, clinics, community centers) with graduate-student coordinators trained in culturally sensitive consent and personal data protection; 4) include deep ancestry measures and descriptive central tendency plus dispersion for all demographic variables; 5) conduct sensitivity analyses that account for clinical and depressive symptom burden and medication-taking status so effect estimates are not confounded by mental-health differences.

Editorial recommendation for authors and editors: require transparent tables that present raw counts, percentages, and recruitment sources, and require that any claims about generalizability explicitly state which ancestries and age groups remain underrepresented and therefore warrant cautious interpretation.

How robust are the replication and cross-cohort validations?

Require replication across at least three independent cohorts: each replication cohort should show the same direction of effect and a replication p-value <0.05 after directionality filtering, with a fixed-effect meta-analytic mean effect size within 1.5-fold of the discovery estimate. If discovery reached genome-wide thresholds, demand replication in cohorts whose combined N is at least two-thirds of the discovery sample or individual cohorts with N>20,000 when trait prevalence is low.

Quantify consistency using I² and leave-one-out: treat I² <40% as acceptable homogeneity; if I²>60% apply random-effects meta-regression including a categorical variable for measurement type, recruitment strategy (early vs later waves), and sex composition (proportion of boys). Run leave-one-out across all cohorts (example: across eleven cohorts when available) to identify single-cohort influence; flag signals that fall away when any one cohort is removed.

Harmonize phenotype definitions before meta-analysis: map cohort instruments (self-report, mother-report, clinical interview) to a common codebook, document items used for measuring partner/gender preference, and perform sensitivity analyses excluding mother-reported or proxy measures. Report subgroup results for boys and for other sex strata; present per-cohort beta, SE, sample N, and recruitment window to allow others to assess selection effects.

Test cross-ancestry replication separately and require consistent direction in at least 75% of ancestry-defined cohorts and nominal significance in at least one non-European cohort when possible. For polygenic scores, require independent R² >0.5% (continuous traits) or OR change >1.05 per SD in independent cohorts; report Nagelkerke R² and the mean change across cohorts rather than a single best-case number.

Address measurement variance explicitly: include a continuous variable coding “measuring method” in meta-regression, model age-at-assessment and early recruitment indicators, and estimate variance explained by cohort-level covariates. Ownership of heterogeneity means publishing cohort-level summaries so ourselves and others can re-weight or exclude cohorts with outlying protocols.

Interpretation guidance: treat small effect sizes linked to behavioral endpoints as biologically plausible but weakly predictive at the individual level; avoid claims that markers predispose a person deterministically. Cross-reference earlier measurement work (savin-williams, díaz, procidano) when mapping items, and state explicit views on limitations of mother vs self reports.

Transparency and reporting checklist: pre-register replication thresholds, publish full summary statistics and analysis code prior to or at time of publishing, include a table of eleven cohort-level metrics (N, sex ratio, mean age, recruitment mode, measurement instrument), and provide a registry of excluded cohorts with reasons so readers can assess potential biases in relationships reported.

What technical limitations (genotyping, phenotype definition) affect interpretation?

Recommendation: require harmonized, pre-registered phenotype definitions and dense genotyping with per-variant INFO >0.8, report ancestry-stratified results and replicate in at least one longitudinal cohort with measurements collected at multiple life stages.

Genotyping limitations: array coverage and imputation drive power and bias. Low-frequency and structural variants are missed by standard arrays; imputation quality falls in non-European ancestries (for example, latino subsets showed increased missingness), producing worse effect estimates and negative bias for rare alleles. Batch effects and per-sample call rates must be reported; exclude variants with call rate <98% or INFO <0.3. Report days between DNA collection and phenotyping when there are temporal gaps: there can be drift in sample handling that affects genotype calling. Polygenic scores built from sparse arrays are often driven by common variants that explain small variance in behaviour-related outcomes; present percent variance explained with confidence intervals.

Phenotype definition and measuring: self-identified labels (self-identified lgbt) differ from behaviour-based measures and romantic preference questions. Harmonize whether the primary measure is sexual behaviour, romantic attraction, identity, or fantasies – each was examined differently across cohorts (ganna and williams used divergent operationalizations). Use both binary and continuous measures where possible; provide z-scored continuous traits and raw counts. Measuring experiences rather than labels reduces misclassification: e.g., number of same-sex partners, age at first romantic experience, or frequency of same-sex days of sexual activity are more reproducible than a single adulthood label.

Timing and life-course: measures collected in adolescence vs adulthood yield different classifications. Longitudinal follow-up reveals fluidity: some participants self-identified at one wave and not at another. Recommend at least two waves separated by months or years; report changes by generation and report proportions with increased, decreased, or stable reporting. Present cross-tabulations for those who changed label and examine whether effect sizes differ for those who remained self-identified across adulthood.

Ascertainment and social context: recruitment via friends, clinics, or online panels biases who participates. Social desirability and stigma create negative measurement error that correlates with age, cohort, and local norms. Include sampling frame details and correct using inverse probability weights or sensitivity analyses. Report subgroup analyses for womens, male, and mixed-sex samples; identify any femalehigh cluster or sex-by-phenotype interactions rather than pooling without stratification.

Analysis reporting: present per-variant INFO, allele frequencies by ancestry, sex-stratified estimates (womens and male), and sensitivity excluding participants with ambiguous or single-wave self-identified labels. Provide effect sizes with standard errors, P-values, and heterogeneity statistics; when small effects are reported, avoid adaptive claims about darwinian selection unless supported by selection scans and functional follow-up. Cite authors by name when comparing methods (for example, ganna reported broad phenotype definitions while williams used more restrictive labels; david and others have examined measurement stability), and release summary statistics that allow others to replicate ancestry- and sex-specific analyses.

Interpretation checklist before claiming population-level relevance: (1) replicate in at least one longitudinal cohort; (2) show results are not driven by batch or recruitment biases; (3) quantify misclassification from self-identified vs behaviour measures; (4) test for heterogeneity by generation and by latino or other ancestry groups; (5) report whether removing participants with unstable labels makes associations stronger or worse. Following these concrete steps reduces misinterpretation and anchors findings in measured experiences rather than labels alone.

Practical implications for clinicians, counselors, and genetic counselors

Begin each encounter with a clear, specific statement: current DNA-based associations are probabilistic and have muy limited predictive value, so no predictive testing for sexual orientation should be offered for children or used as a sole basis for clinical decisions.

Use concrete figures when explaining risk: report that polygenic scores explain only a small fraction of variance (single-digit percentage points in most large analyses), that effect sizes vary by sample design and region, and that predictive scores trained on primarily white samples transfer poorly across race and are not clinically actionable. Cite professional guidance such as the American Psychological Association for ethical context: https://www.apa.org/topics/lgbtq/orientation.

Adopt a structured intake that captures behavior-based history with validated tools (e.g., Klein/Kinsey-style scales, sexual history checklists, and social support measures such as Procidano instruments) so clinicians can separate identity, behavior, and mental health needs. Record inventory results and symptom scores, and track change over days or weeks when monitoring therapy response.

When communicating research findings, avoid deterministic language; instead explain that most genetic components examined so far are probabilistic and contextual. An editorial on prior work noted that non-representative design helped bias to persist; clinicians should therefore comment on sampling limits when patients cite articles. Explain that some publications examined regional or northeast samples, womens cohorts, or those that took convenience recruitment, which limits generalizability.

Assess minority stress and exposure to anti-gay stigma explicitly: screen for experiences of discrimination, coming-out challenges, and internalized homophobia or biphobia. Offer referrals for mental health care when screening indicates reduced functioning. Use behavior-based screening items alongside mental health measures to identify central drivers of distress rather than attributing symptoms to biology alone.

For genetic counselors: include a scripted explanation of what a polygenic score does and does not mean, include family history examining relevant components, and document informed consent that patients sought testing after being helped to understand limitations. Recommend deferral of predictive testing for orientation-related traits; discuss potential harms such as discrimination, insurance impact, and misuse by others.

Clinical teams should adopt a representative-data approach when interpreting new findings: flag samples that are predominantly white or region-specific, note if bi-attraction subgroups were underpowered, and treat subgroup results as hypothesis-generating rather than conclusive. Klesse and others have examined social dimensions that interact with biological components; include social context when formulating care plans.

Operational recommendations: (1) add a short consent script to electronic records explaining low predictive value; (2) include a documentation checkbox that counseling covered social, legal, and medical implications; (3) use validated inventories and behavior-based measures at baseline and follow-up; (4) create a referral list for community supports such as local ball organizers and peer groups, which have helped build resilience.

Monitor for bias in care: audit charts quarterly for language that frames orientation as deterministic, track rates of referrals to mental health vs genetics, and provide staff training that reduces ignorance and anti-gay assumptions. These steps produced reduced miscommunication in pilot clinics and were described as a useful, practical component of clinician education in a recent article on clinical translation of genomics in medicine.

Keep communications concise and patient-centered: acknowledge unanswered questions, persist in explaining uncertainty, and emphasize that identity, behavior, and attraction can coexist within varied patterns (including bi-attraction). Such an approach fosters trust, supports autonomy, and is a more ideal clinical stance than speculative prediction.

Should genetic findings change intake questions or family histories?

Recommendation: Do not add biomarker-specific items to universal intake; instead deploy an optional, consented family-history module that captures parental composition, adoption status, maternal and paternal histories, and relevant cognitive and mental-health measures, with clear limits on clinical use.

• Rationale: regression analyses from recent papers show that biological associations often denote higher odds for some outcomes but account for a small proportion of variance; many effects are reduced or vanish once demographic covariates and family structure are accounted.
• When to include: add the module only for patients or research participants who are explicitly interested and have provided written consent, or when the information will change clinical management or health risk assessment.
• Privacy and storage: store responses separately, limit access, and document purpose; do not merge optional responses into default EMR fields without patient consent because that increases privacy risk.
• Analysis standards: require regression models that adjust for ancestry, parental ages, socioeconomic measures and family clustering; report variance accounted and avoid overinterpretation of small effects.

Artículos de ingesta sugeridos (módulo opcional):

1. Composición familiar: “¿Eres adoptado/a? (sí/no)”; “¿Estuvo presente un padre biológico en la infancia? (presente/ausente/desconocido)”.
2. Identidades parentales (opcional): “Si lo desea, indique las identidades auto-descritas de los padres (opciones: masculino, femenino, transgénero, lesbiana, gay, bisexual, otro).” – incluir exclusión explícita.
3. Salud familiar: “Antecedentes de trastornos psiquiátricos, cognitivos o del desarrollo en familiares de primer o segundo grado (enumerar y edad de inicio)”. – incluir medidas como dificultades de aprendizaje, TDAH, trastornos del estado de ánimo importantes.
4. Exposiciones a lo largo de la vida: “¿Alguna exposición prenatal conocida, enfermedad materna durante el embarazo o interrupciones en el cuidado de la primera infancia?”
5. Consentimiento para la investigación: “¿Consiente a la vinculación de este módulo de antecedentes familiares con conjuntos de datos de investigación? (sí/no)”.

Lista de verificación operativa para clínicas y equipos de investigación:

• Borrador del lenguaje de consentimiento que explique por qué se recopilan los elementos, cómo se utilizarán los resultados y que las respuestas no afectarán la atención a menos que sean clínicamente relevantes.
• Capacitar al personal de admisión para evitar suposiciones y utilizar frases neutrales; realizar juegos de roles con escenarios (ejemplo: David, cuyo padre estuvo ausente y que fue adoptado, no debe ser presuponer que tiene características particulares).
• Indexar respuestas opcionales en el EMR con etiquetas de acceso restringido y registros de auditoría; no exponer módulos a portales de facturación o externos sin re-consentimiento.
• Exigir que las canalizaciones analíticas presenten los tamaños del efecto, los intervalos de confianza y el porcentaje de varianza explicado en lugar de etiquetas binarias; utilizar diagnósticos de regresión para probar la robustez.
• Publicar una breve FAQ para pacientes que explique la relación entre los elementos del historial familiar y la salud, enfatizando la diversidad y los límites del valor predictivo.

Ejemplos de viñetas que denotan un uso apropiado:

• Caso A: La clínica de Johns agrega el módulo para un grupo de investigación; los resultados de regresión muestran una tasa más alta de un informe específico, pero el efecto se reduce después de tener en cuenta las medidas socioeconómicas y la edad materna – resultado del informe con el porcentaje de varianza explicado.
• Caso B: El equipo de Patel utiliza el módulo cuando un paciente está interesado en el asesoramiento sobre reproducción y documenta los antecedentes familiares maternos y paternos que informan de manera significativa las decisiones de atención.
• Evitar equipos clínicos que recopilen datos de identificación de los padres sin consentimiento o justificación clínica; evitar preguntas rutinarias sobre parejas del mismo sexo o rasgos privados a menos que sean directamente relevantes para la atención.

Vista final: las preguntas sobre la anamnesis y el historial familiar deben enfatizar la elección del paciente, el consentimiento transparente, la elaboración analítica rigurosa (basada en regresión, varianza explicada) y el respeto por la diversidad; recolectar información sobre la madre, el padre, la adopción y los padres ausentes cuando esté directamente relacionada con la salud o cuando el paciente esté explícitamente interesado.

¿Cuándo discutir los resultados de estudios genéticos con clientes o pacientes?

Discuta los resultados rápidamente cuando un hallazgo cambiará el manejo clínico o cuando el paciente solicite divulgación: para una variante particular que altere la selección de medicamentos, programe una discusión cara a cara o por video dentro de los siete días calendario, documente el consentimiento informado y aclare la propiedad de los datos.

Si los investigadores tomaron una muestra de menores o grupos vulnerables, involucre a los tutores: cuando las muestras provinieron de adolescentes o jóvenes, proporcione procesos de notificación y asentimiento del tutor; para los participantes de nivel de graduado, trate el consentimiento como de nivel adulto pero confirme la comprensión del encuestado y cualquier preferencia previa registrada.

Contextualizar los resultados numéricos: informar el puntaje bruto, el percentil y la dirección, indicar cómo se desempeñó el resultado en relación con los controles emparejados y presentar comparaciones de subgrupos (ejemplos: muestra regional del oeste, subgrupo latino, cohorte de Michigan) para mostrar diferentes riesgos de línea base en comparación con la población general.

Si un resultado se asocia con un mayor riesgo o un nuevo síntoma, describe pasos de seguimiento concretos: establece intervalos de monitoreo, enumera señales de alerta, proporciona asesoramiento adicional y referencias específicas que puedan ayudar, documenta quién elaboró el plan de atención y cuándo, y evita hacer predicciones definitivas sobre el comportamiento.

Proporcionar contexto científico y opciones de comunicación: explicar la incertidumbre, describir la fluidez esperada de la evidencia con el tiempo, ofrecer citas y posibilidad de recontacto si se desea una reanálisis, indicar un contacto de la clínica como Williams para el seguimiento, y explicar qué podría ocurrir si la evidencia o las recomendaciones cambian y quién sugerirá los próximos pasos.