Stratifying patients using fast multiple kernel learning framework: case studies of Alzheimer's disease and cancers

Area of Science:

• Biomedical data analysis
• Machine learning in healthcare
• Computational biology

Background:

• Predictive patient stratification is crucial for personalized medicine, identifying patients who will benefit from specific interventions.
• Integrating heterogeneous and high-dimensional biomedical data presents significant computational and machine learning challenges.
• Existing methods struggle with vast dimensional space handling and data type disparities in machine learning for biomedical data.

Purpose of the Study:

• To develop an efficient machine learning framework for handling complex biomedical data integration and stratification.
• To propose a fast multiple kernel learning framework (fMKL-DR) for optimizing matrix chain multiplication and data dimension reduction.
• To validate the fMKL-DR framework in real-world case studies for Alzheimer's disease and cancer patient stratification.

Main Methods:

• Developed a fast multiple kernel learning framework (fMKL-DR) optimizing matrix chain multiplication for dimension reduction.
• Applied the fMKL-DR framework to Alzheimer's disease (AD) patient stratification using MRI data.
• Utilized the fMKL-DR framework for stratifying six types of cancer by integrating gene expression, miRNA expression, and DNA methylation data.

Main Results:

• Achieved high accuracy (AUC close to 1) in classifying AD patients and different disease phases using MRI data.
• Demonstrated significantly improved classification accuracy for cancer patients by integrating multiple data types compared to single data types.
• The fMKL-DR framework successfully integrated gene expression, miRNA expression, and DNA methylation for cancer stratification.

Conclusions:

• The fMKL-DR framework significantly reduces computational cost and enhances accuracy in AD and cancer patient stratification.
• Optimized data integration, dimension reduction, and kernel fusion processes within the framework.
• fMKL-DR shows great potential for analyzing large-scale cohort data and advancing personalized prevention strategies.