MKpLMM

MKpLMM implements a multi-kernel penalized linear mixed model with adaptive lasso to integrate multi-omics data for high-dimensional genomic and phenotype risk prediction.

Key Features:

• Multi-Kernel Approach: MKpLMM utilizes multiple kernel functions to capture predictive effects and interactions across different omics data layers.
• Adaptive Lasso Penalization: It applies adaptive lasso penalization for variable selection, enabling robust identification of predictive regions and layers.
• Extension of Linear Mixed Models: The method extends standard linear mixed models to accommodate high-dimensional multi-omics data through kernel integration and penalization.
• Data-Driven Selection: MKpLMM employs a data-driven procedure to select predictive genomic regions and omics layers from high-dimensional datasets.

Scientific Applications:

• Phenotype Prediction: Demonstrated in simulation studies and real-data analyses to outperform competing methods in phenotype prediction tasks.
• PET-Imaging Outcomes Analysis: Applied to analyze PET-imaging outcomes in Alzheimer's Disease Neuroimaging Initiative studies.
• Drug Response Prediction: Used to predict drug responses across 64 different cases for personalized medicine analyses.

Methodology:

MKpLMM extends standard linear mixed models by integrating multi-kernel functions with adaptive lasso penalization to capture individual and interactive effects across omics layers and employs data-driven selection of predictive regions and layers.