Multilevel Survival Modeling With Structured Penalties for Disease Prediction From Imaging Genetics Data

Area of Science:

• Biomedical data analysis
• Computational biology
• Medical imaging and genetics

Background:

• Accurate disease prediction is crucial for early intervention, especially for neurodegenerative diseases like Alzheimer's.
• Existing models often fail to adequately integrate multimodal data (genetic and imaging) or account for disease progression over time.
• Classical additive models may discard valuable information from certain data modalities if their contributions are unbalanced.

Purpose of the Study:

• To introduce a novel multilevel survival model for predicting the timing of disease onset.
• To effectively model interactions between genetic and imaging data for enhanced prediction accuracy.
• To overcome limitations of classification-based approaches and fixed time frames in disease prediction.

Main Methods:

• Development of a multilevel survival model incorporating genetic (SNP) and imaging (MRI) data.
• Application of tailored penalties: group lasso for genetic data and L2 penalty for imaging data.
• Implementation of a fast proximal gradient-based optimization algorithm.

Main Results:

• The model accurately predicted the time to Alzheimer's disease (AD) conversion in patients with mild cognitive impairment (MCI).
• Demonstrated significant interactions between genetic variants and brain imaging alterations in predicting disease status.
• Validated on the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.

Conclusions:

• The proposed multilevel survival model offers a powerful framework for disease prediction using multimodal data.
• The method effectively captures complex interactions between genetic and imaging biomarkers.
• This generic approach holds potential for predicting various other diseases beyond Alzheimer's.