A sparse PLS for variable selection when integrating omics data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics
• Proteomics
• Metabolomics

Background:

• Advances in biotechnology enable the integration of diverse omics data (e.g., transcriptomic, proteomic, metabolomic).
• Traditional feature selection methods are insufficient for integrated omics data analysis.
• Biologists require tools for interpreting complex, high-dimensional datasets.

Purpose of the Study:

• To develop a one-step procedure for integrating two-block omics data.
• To achieve simultaneous variable selection from integrated datasets.
• To facilitate biological interpretation through a novel computational method.

Main Methods:

• Introduction of a novel computational methodology: sparse Partial Least Squares (PLS).
• Integration of two-block data measured on the same samples.
• Application of Lasso penalization on PLS loading vectors during Singular Value Decomposition for sparsity.

Main Results:

• Sparse PLS demonstrates effectiveness in predictive analysis and variable selection for integrated omics data.
• The method yields biologically meaningful results, as shown in comparisons with classical PLS.
• A thorough biological interpretation was provided for one real-world dataset.

Conclusions:

• Sparse PLS is a valuable tool for variable selection in high-dimensional, integrated omics datasets.
• The one-step integration and selection procedure simplifies data analysis for biologists.
• This approach enhances the interpretability of complex biological data.