Multilevel heterogeneous omics data integration with kernel fusion.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput omics data generation is rapidly increasing.
• Integrating diverse omics data is crucial for understanding complex diseases.
• Supervised learning models with predefined structures struggle with unknown feature relationships in omics data.

Purpose of the Study:

• To propose a novel fused kernel partial least squares (fKPLS) model for disease classification and prediction using multilevel omics data.
• To address effect heterogeneity where different omics data types contribute differently to disease traits.
• To enhance prediction performance by effectively integrating multiple omics data types.

Main Methods:

• Reviewed kernel fusion (KF) methods based on support vector machine and kernel partial least squares (KPLS).
• Developed a fused KPLS (fKPLS) model incorporating a fused kernel to handle effect heterogeneity.
• Optimized kernel parameters and weights using a genetic algorithm (GA), resulting in the GA-fKPLS model.

Main Results:

• The proposed GA-fKPLS model substantially improves disease classification performance.
• Demonstrated effectiveness through extensive simulations and real-world data analysis.
• The fused kernel approach effectively integrates multiple omics data types for better prediction.

Conclusions:

• The GA-fKPLS model offers a powerful approach for disease classification by integrating multilevel omics data.
• The kernel fusion method can be extended to other kernel-based analyses, including association studies.
• This work provides a robust framework for leveraging omics data in disease research.