Sequential Pathway Inference for Multimodal Neuroimaging Analysis.

Area of Science:

• Neuroscience
• Biostatistics
• Genomics

Background:

• Sequential mediation analysis is vital for understanding complex disease pathways, particularly in Alzheimer's disease research.
• Existing methods often overlook mediator dependency or are unsuitable for sequential, multimodal data.
• Hypothesis testing in sequential mediation presents unique challenges compared to sparse estimation.

Purpose of the Study:

• To develop a statistical inference procedure for hypothesis testing in sequential mediation analysis with multiple, ordered data modalities.
• To address the limitations of existing methods by accounting for conditional dependency among mediators.
• To enable robust testing of mediation pathways in complex, multimodal datasets.

Main Methods:

• Proposed a novel statistical inference procedure for sequential mediation analysis.
• Developed methods to handle conditionally dependent mediators within and across modalities.
• Allowed the number of mediators to grow with sample size.
• Established theoretical guarantees for significance quantification, including asymptotic size, power, and false discovery control.

Main Results:

• The proposed method effectively quantifies significance in sequential mediation pathways.
• Theoretical guarantees ensure reliable performance in terms of statistical power and error control.
• Demonstrated the method's efficacy through comprehensive simulations.
• Successfully applied the method to a real-world multimodal neuroimaging dataset for Alzheimer's disease.

Conclusions:

• The developed statistical procedure provides a robust framework for hypothesis testing in sequential mediation analysis.
• The method is applicable to complex, multimodal data, accommodating dependent mediators.
• This work advances the analysis of neuroimaging and other multimodal data for diseases like Alzheimer's.