Researchers introduced ALADYNOULLI, a Bayesian generative model that integrates electronic health records and genetic data to reveal underlying disease signatures, according to a study published in Nature on Jul. 16. The framework was applied across three major biobanks and traced how interconnected disease risks change over the life course, exposing differences hidden within conventional diagnoses.
The study evaluated ALADYNOULLI's performance in the United Kingdom Biobank, showing stronger discrimination of short- and long-term risk than established clinical risk scores. The model predicted disease-level phenotypes rather than individual diagnostic codes. Researchers said that following prospective validation, the framework could help identify individuals who may benefit from closer preventive assessment.
ALADYNOULLI modeled age and EHR diagnoses along with polygenic risk scores to reveal temporal patterns in disease risk among diagnostic subgroups. The team applied the model to datasets from All of Us, Mass General Brigham, and United Kingdom Biobank—covering more than 683,000 participants with records spanning up to 52 years—and performed signature-based genome-wide association studies as well as rare variant association studies.
The researchers used PheCodes for grouping related ICD-10 codes into broader phenotypes for risk prediction. Inverse probability weighting addressed potential participation bias in UKB while preserving core relationships between diseases and signatures. Model performance was compared against established clinical scores such as the Pooled Cohort Equation for cardiovascular disease and the Gail model for breast cancer over both one-year short-term and ten-year long-term periods.
With 21 latent components representing distinct disease signatures or reference groups, ALADYNOULLI showed high cross-cohort preservation of signature composition (median 80%) and identified subgroups within broader categories reflecting known biological characteristics. For example, individuals with familial hypercholesterolaemia-associated variants showed enrichment of cardiovascular signatures; those with clonal haematopoiesis displayed stronger inflammatory signatures; rare variant burdens aligned with expected patterns for LDLR, TTN, or BRCA2 genes; higher polygenic risk scores corresponded to stronger relevant signatures.
The authors concluded that by integrating longitudinal EHRs and genetic data, their framework estimates how disease risks change across the lifespan, but noted limitations due to incomplete histories or unmodeled environmental factors. They said further external validation is needed before supporting personalized medicine approaches.
