EXPLAINABLE MULTI-OMICS AI FOR PRECISION RISK STRATIFICATION OF INHERITED DISORDERS IN PAKISTANI FAMILIES

Abstract

Inherited disorders present substantial challenges for accurate diagnosis and individualized risk assessment, particularly in genetically diverse and highly consanguineous populations. This study examined an explainable multi-omics artificial intelligence (AI) framework for precision risk stratification of inherited disorders among Pakistani families. A quantitative predictive design was employed using integrated genomic, transcriptomic, proteomic, metabolomic, familial, and clinical indicators. Multiple machine-learning algorithms were compared using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC), while SHAP-based explainable AI was applied to identify the principal contributors to individual risk predictions. The analysis indicated that Gradient Boosting achieved the strongest predictive performance (accuracy = 87.6%; AUC = .94), compared with genomic information alone (accuracy = 78.9%; AUC = .83). Integration of clinical and multi-omics information increased the AUC from .83 to .94. Pathogenic genomic variants, multi-omics risk scores, clinical phenotype severity, family history, and consanguinity emerged as major predictive features. The findings suggest that integrating heterogeneous molecular and clinical information with explainable AI can provide a more comprehensive and interpretable approach to inherited-disorder risk stratification. The framework may support precision diagnosis, genetic counselling, family screening, and individualized clinical decision-making in Pakistan. However, the reported numerical results require validation using real-world, longitudinal, and independently collected Pakistani datasets.