Sparse partial least squares with group and subgroup structure.

Area of Science:

• Biostatistics
• Genomics
• Immunology

Background:

• Integrative analysis of high-dimensional omics data is crucial for understanding complex biological systems.
• Prior knowledge of variable relationships enhances the elucidation of connections between different omics datasets.
• Identifying relationships between genomic expression and cytokine data is vital for human immunodeficiency virus (HIV) vaccine research.

Purpose of the Study:

• To identify significant relationships between genomic expression and cytokine data within an HIV vaccine trial.
• To propose a novel statistical method for integrative omics analysis that accounts for biological structures.
• To generalize and connect existing sparse modeling techniques in partial least squares (PLS).

Main Methods:

• Development of a flexible partial least squares (PLS) technique incorporating group and subgroup structures.
• The method accounts for the grouping of genetic markers (e.g., gene sets) and temporal effects.
• Theoretical connections established to variable selection methods for both supervised and unsupervised problems.

Main Results:

• Simulation studies demonstrated the method's performance compared to alternative sparse approaches.
• The proposed technique successfully integrates genomic and cytokine data, considering biological structures.
• The method offers a generalized framework for sparse PLS, enhancing variable selection capabilities.

Conclusions:

• The novel flexible partial least squares method provides a powerful tool for integrative omics analysis in HIV research.
• The approach effectively handles grouped and temporal omics data, improving relationship discovery.
• The R package 'sgspls' is available for researchers to apply this advanced statistical methodology.