OPTIMUS: Predicting Multivariate Outcomes in Alzheimer's Disease Using Multi-modal Data amidst Missing Values

Multimodal data improves disease prediction: The study demonstrates that combining multiple types of data (neural, genetic, and proteomic) provides better predictions of Alzheimer's disease outcomes than traditional single-modality approaches, suggesting that a more comprehensive view of the disease leads to more accurate forecasting.

Missing data is a practical challenge that can be addressed: Rather than discarding incomplete datasets, OPTIMUS uses modality-specific imputation techniques to recover missing information while maintaining prediction accuracy, making the framework applicable to real-world medical data that is often incomplete.

Different biomarkers predict different disease aspects: The research reveals that neural and transcriptomic signatures have differential predictive power for various outcomes, meaning that predicting cognitive decline may require different biological markers than predicting behavioral changes—an important distinction for personalized medicine.

Explainability is essential for clinical adoption: By incorporating explainable AI (XAI) methods, OPTIMUS makes its predictions interpretable to clinicians and researchers, addressing a critical gap where "black box" machine learning models may be rejected in healthcare settings despite superior accuracy.

Large-scale validation across disease stages: The study uses a substantial dataset spanning 346 cognitively normal subjects, 608 with mild cognitive impairment, and 251 AD patients, providing evidence across the disease spectrum rather than isolated patient groups.