Alzheimer’s disease (AD) has a long preclinical phase, yet current diagnostic approaches typically identify patients only after substantial neurodegeneration, limiting opportunities for prevention. Blood-based biomarkers provide a scalable and minimally invasive strategy for earlier risk detection. This project aims to determine whether longitudinal plasma protein trajectories can predict incident AD, with a specific focus on sex-specific vulnerability. Women are disproportionately affected by AD, particularly during and after the menopausal transition, but female -specific proteomic signatures remain remain poorly characterised.
Using UK Biobank proteomics (~50,000 participants), we will: (i) identify plasma proteins associated with incident AD risk using Cox proportional hazards models, supplemented by causal machine learning methods (e.g., causal random forest, gradient boosting, SVM) to strengthen inference beyond prediction; (ii) characterise longitudinal protein trajectories stratified by sex and menopausal status; (iii) integrate proteomic signals with neuroimaging (MRI-derived cortical atrophy) and genetic risk (APOE genotype, polygenic risk scores); (iv) apply Mendelian randomisation (leveraging pQTL data) to assess causal links between proteins and AD risk; and (v) construct interpretable machine learning models (XGBoost + SHAP) for multimodal risk prediction.
Results will be validated in ADNI and other complementary open datasets. Key deliverables include a female-specific biomarker panel, trajectory maps delineating proteomic changes preceding diagnosis, and interpretable risk models for early prediction. By integrating causal inference with multi-modal data, this project aims not only to improve prediction accuracy but also to elucidate inflammatory and hormonal pathways contributing to cortical vulnerability in women. The ultimate goal is to inform sex-specific strategies for early detection and precision prevention of AD.