Multi-kernel linear mixed model with adaptive lasso for prediction analysis on high-dimensional multi-omics data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Precision medicine relies on accurate risk prediction models using genomic and other factors.
• High-dimensional multi-omics data offer rich resources but lack developed analytical methods.

Purpose of the Study:

• To develop an advanced predictive modeling framework for multi-omics data analysis.
• To enhance genomic risk prediction by integrating diverse omics data layers and their interactions.

Main Methods:

• Introduced a multi-kernel penalized linear mixed model with adaptive lasso (MKpLMM).
• MKpLMM extends standard linear mixed models to capture individual omics layer effects and interactions.
• Employed a data-driven approach for selecting predictive omics regions and layers.

Main Results:

• MKpLMM demonstrated superior performance in phenotype prediction across simulations and real-world datasets.
• The model accurately predicted PET-imaging outcomes in Alzheimer's Disease Neuroimaging Initiative study.
• MKpLMM showed strong predictive capabilities for 64 drug responses.

Conclusions:

• MKpLMM provides a robust and effective framework for multi-omics data analysis in precision medicine.
• The developed method outperforms existing approaches in genomic risk prediction.
• The R-package OmicPred is available for broader application.