OPTIMUS: Predicting Multivariate Outcomes in Alzheimer's Disease Using Multi-Modal Data Amidst Missing Values.

Area of Science:

• Neuroscience
• Genetics
• Computational Biology
• Artificial Intelligence

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder influenced by neural, genetic, and proteomic factors.
• Traditional AD prediction often uses univariate outcomes, overlooking the complexity of multimodal data and missing measurements.
• Existing machine learning (ML) approaches show promise but lack biological interpretability.

Purpose of the Study:

• To develop a predictive and explainable ML framework (OPTIMUS) for analyzing multimodal data in AD.
• To uncover many-to-many predictive relationships between multimodal data and multivariate AD outcomes, addressing missing data.
• To enhance the biological understanding of AD through interpretable ML models.

Main Methods:

• Developed OPTIMUS, integrating missing data analysis, predictive modeling, multimodal data analysis, and explainable AI (XAI).
• Applied imputation techniques to handle missing data within modalities while optimizing prediction.
• Utilized ML to map multimodal biomarkers to multivariate outcomes and XAI for biomarker explanation.

Main Results:

• OPTIMUS successfully handled missing data and improved prediction accuracy.
• Identified specific neural and transcriptomic signatures predictive of memory, executive function, visuospatial function, and language outcomes.
• Demonstrated joint but differential prediction of outcomes by identified biomarkers.

Conclusions:

• OPTIMUS provides a framework for building predictive and biologically explainable ML models for AD.
• The study highlights the potential of multimodal biomarkers in capturing diverse AD disease profiles.
• Findings advance the understanding of complex, many-to-many pathways in Alzheimer's disease.