Disease causal pathways play an integral role in developing strategies for disease prevention and intervention. To elucidate these pathways, utilizing large-scale datasets that integrate multi-omics data have proven to be useful, especially for gaining new insights into disease causal mechanisms. However, there are still challenges to address, specifically with complex diseases that possess multiple subtypes with distinct pathological processes (e.g., dementia). Furthermore, existing predictive models typically rely on associative learning and selective clinical knowledge, often missing important or under-researched risk factors. Causal models, in contrast, offer a framework to discover new pathways and determine the causal effects between exposures and outcomes. A research gap exists in robust and scalable causal models that properly account for statistical constraints and assumptions. Rather than focusing on a single condition, this project’s primary objective is to develop a causal discovery method that is broadly applicable to a variety of aging-related health conditions. Both large-scale data and prior domain knowledge will be integrated into a generalisable and scalable causal model. We seek to address three key questions: 1) Pathway Identification: Can we identify the underlying pathways leading to a disease or its subtypes? 2) Prevention Targets: What key conditions or factors serve as precursory targets for prevention? 3) Impactful Interventions: Which modifiable factors can significantly alter the course of the disease in individuals? To address these research questions, we will combine methods such as Bayesian networks and Mendelian randomization for causal discovery and estimation. We will leverage genomic, lifestyle, and biological data from the UK Biobank to build and train a novel causal model adaptable to a variety of diseases. To ensure robustness and generalisability, the final model will be rigorously validated using independent datasets.
