Genomic Biobank Studies Advance Personalized Disease Prediction and Psychiatric Risk Discovery

Recent studies leveraging genomic data from large biobanks are advancing personalized prediction of disease and deepening understanding of shared genetic liability between psychiatric and physical conditions. One study demonstrated a practical recall-by-genotype strategy for precision psychiatry, another identified pervasive genetic correlations across psychiatric and physical illness systems, and a third found that integrating disease-history trajectory information improves prediction performance.

In a study published in npj Genomic Medicine, researchers from the Icahn School of Medicine at Mount Sinai used BioMe, one of the nation's largest and most diverse healthcare system biobanks, to identify individuals carrying rare copy number variants (CNVs) that substantially increase risk of neurodevelopmental disorders including autism spectrum disorder, intellectual disability, and schizophrenia. They recontacted 892 participants — 335 CNV carriers, 217 individuals with schizophrenia without these variants, and 340 neurotypical controls — to evaluate whether recall-by-genotype could be implemented. Eighteen percent of participants responded to recruitment and 8 percent completed comprehensive psychiatric and cognitive assessments. The final cohort was diverse, with participants self-identifying as 37 percent African ancestry, 34 percent Hispanic, and 26 percent European ancestry. Detailed evaluations identified developmental, clinical, and cognitive characteristics beyond those captured in routine electronic health records. The study establishes operational benchmarks for implementing recall-by-genotype studies within diverse healthcare systems. The senior author stated, "By recontacting participants carrying rare CNVs for detailed assessments, we demonstrated both opportunities and challenges of the recall-by-genotype study design in a large, diverse healthcare system biobank."

Another study, applying Genomic SEM and introducing Genomic E-SEM to 73 physical outcomes (~1.9 million cases) across eight medical domains, modeled latent physical illness factors alongside psychiatric factors. Thought/psychotic and compulsive factors exhibited few associations with physical illness factors, whereas internalizing, neurodevelopmental, and substance-use factors showed substantial genome-wide genetic correlations with all physical illness systems. A multivariate GWAS of a transdiagnostic physical illness factor defined by 21 physical disorders identified 27 genomic risk loci and exhibited substantial predictive validity in subsequent PheWAS. The findings reveal pervasive risk sharing between specific groups of psychiatric and physical conditions and suggest the need for additional nosological frameworks.

A third study proposed a personalized disease prediction framework integrating structured genomic variant annotations, deep semantic embeddings of longitudinal disease histories, and process conformance metrics derived from historical disease pathways. Disease history embeddings were generated using three BERT-based models — BioBERT, BioClinicalBERT, and BiomedBERT — and alignment-based process conformance checking measured how closely an individual's disease trajectory conforms to typical population progression patterns. Across seven feature configurations, incorporating conformance-based fitness features improved prediction performance across all disease categories and classifiers, yielding higher AUROC values and lower Brier scores, while embedding-only and genomic variant annotation-only configurations ranked among the lowest-performing models. The findings indicate that process-level disease pathway conformity captures temporal and behavioral information not fully represented by genomic or deep semantic features alone, and future work will target more narrowly defined diseases.