PHBC estimates shared genetic variance across diseases: 48% of SNP heritability with an expanded auxiliary trait set
Background: Beyond genetic correlations between disease pairs
A study of 15 diseases in the UK Biobank estimated that, on average, 27% of heritability attributable to common single-nucleotide polymorphisms (SNPs) was shared with other diseases. The estimate was 48% when the set of auxiliary diseases and traits was expanded. Published in Nature Genetics, the study proposes PHBC, a bias-corrected method for estimating this pleiotropically shared proportion of heritability.
Pleiotropy occurs when genetic factors affect multiple traits. Earlier studies have focused mainly on estimating genetic correlations between pairs of diseases or identifying genetic components common to several diseases. How much of a given disease’s genetic variance is shared with multiple diseases and traits? Because the information shared among diseases overlaps, simply adding individual genetic correlations cannot answer this question.
Key findings: Correcting estimation noise lowered the estimate from 38% to 27%
PHBC estimates shared genetic variance from a genetic correlation matrix derived from genome-wide association study (GWAS) summary statistics. With finite samples, estimates retain noise even when no true genetic correlation exists, potentially inflating the estimated shared proportion. The researchers corrected this bias using a Monte Carlo approach involving repeated random sampling. In simulations, the corrected estimates were generally close to the true values specified in the simulations.
The analysis of real data used GWAS summary statistics for 15 diseases across seven categories, derived from genetic data on 157,206 unrelated people of British ancestry. The mean proportion of shared SNP heritability across diseases fell from 38% (standard error, 4 percentage points) before correction to 27% (standard error, 3 percentage points) after correction. Whether the estimates were corrected substantially affected their interpretation. In the same analysis, the estimates were 44% (standard error, 8 percentage points) for major depressive disorder and 45% (standard error, 6 percentage points) for type 2 diabetes.
The team expanded the auxiliary set by adding 30 diseases and 17 quantitative traits from public GWAS meta-analyses. Some of those meta-analyses also included UK Biobank data. With an auxiliary set comprising 45 diseases and 17 quantitative traits, the mean shared proportion for the original 15 target diseases was 48% (standard error, 5 percentage points). Each analysis used auxiliary diseases and traits whose squared genetic correlation with the target disease was below 0.5.
In the primary analysis, excluding auxiliary diseases from the same category as the target disease lowered the mean only from 27% to 24% (standard error, 2 percentage points). This indicates that shared components extend across disease categories rather than being concentrated only within the same category.
Implications and outlook: The denominator and scope of the shared proportion
The 48% figure does not represent an individual’s probability of developing a disease or half of all causes of disease. It is the proportion of genetic variance explained by the analyzed common SNPs that can be explained by combining genetic components of the selected auxiliary diseases and traits. It is also distinct from a value obtained by identifying individual pleiotropic loci and summing all their contributions. The quantity being estimated changes with the auxiliary trait set and the cohort’s disease definitions.
The researchers suggested that the results could inform the construction of prior information in polygenic risk score studies, or that the approach could be extended to analyze shared components by tissue, cell type, and gene pathway. The study did not test improvements in patient prognosis or effects on preventing comorbidities. Because the data were predominantly from people of European ancestry, validation in other populations remains a task for future research.
The authors also noted a limitation: the stochasticity of the process used to remove highly genetically correlated auxiliary traits to ensure numerical stability was not reflected in the reported standard errors. Estimates of shared genetic variance should therefore be interpreted in light of this estimation uncertainty and the conditions of the analysis.
The methods and results are available in the Nature Genetics article.
Nature Genetics, Published online: 09 June 2026. DOI: 10.1038/s41588-026-02607-w
Summary: Bypassing the low prediction velocities and summary statistic inflation errors that historically cloud empirical cross-trait linkage disequilibrium score regression in complex comorbidities, this multi-omic translation scales a programmable statistical mapping infrastructure termed PHBC. Utilizing mathematical bias-correction frameworks synchronized with deep-depth phenotypic registers across the UK Biobank database, the computing platform systematically charts the pleiotropic shared heritability architecture of target diseases against extensive auxiliary disease sets concurrently. This molecular calibration delivers a validated, non-invasive computational baseline to isolate pervasive sharing mechanics across discrete disease categories, suppress sample-overlap artifacts, and guide prospective adaptive cohort stratification under precision multi-trait genomic governance.
Researchers interpreting GWAS data across multiple diseases can use PHBC to estimate how much of a target disease’s genetic variance attributable to common SNPs is shared with other diseases and traits. For example, when analyzing type 2 diabetes alongside other diseases, the method can help reduce overestimation due to noise in genetic correlation estimates and compare how the shared proportion changes with the auxiliary trait set.
These results may inform the construction of prior information for polygenic risk score studies that combine information across traits. If methods are developed to break down shared components by tissue, cell type, or gene pathway, the approach could also be extended to research assessing priorities among drug targets associated with multiple diseases. This study did not demonstrate improved individual prediction or treatment effects, so practical application would require separate validation.