Novel feature selection method via kernel tensor decomposition for improved multi-omics data analysis.

Area of Science:

• Computational biology
• Bioinformatics
• Data science

Background:

• Multi-omics data analysis presents challenges in feature selection due to large dataset sizes.
• Determining appropriate weighting for individual omics datasets is unclear and impacts feature selection outcomes.
• Existing methods for integrating multi-omics data often lack clarity in weighting strategies.

Purpose of the Study:

• To extend a kernel tensor decomposition (KTD)-based unsupervised feature extraction (FE) method for multi-omics data integration.
• To develop a weight-free approach for integrating multi-omics datasets from common samples.
• To improve the efficiency and interpretability of feature selection in multi-omics analysis.

Main Methods:

• Kernel tensor decomposition (KTD)-based unsupervised feature extraction (FE) was reformatted.
• The method was applied to synthetic and real multi-omics datasets.
• Kernelized tensors sharing common samples were utilized.

Main Results:

• The advanced KTD method demonstrated comparable performance to existing tensor decomposition methods.
• The proposed method required reduced memory and central processing unit (CPU) time.
• This KTD method assigns P values to features, a rare capability in multi-omics analysis.

Conclusions:

• The advanced KTD method is effective for weight-free multi-omics data integration and feature selection.
• The method offers computational advantages and enhanced interpretability through P-value assignment.
• The developed R code is available for reproducible research.