Group and sparse group partial least square approaches applied in genomics context.

Area of Science:

• Computational biology
• Statistical genetics
• Bioinformatics

Background:

• Integrating large, complex 'omics' datasets presents significant computational challenges.
• Existing sparse partial least square (sPLS) methods do not account for biological pathway structures or marker group effects.
• Incorporating predefined marker groups can enhance the relevance and efficacy of multivariate statistical approaches.

Purpose of the Study:

• To extend partial least square (PLS) methods by incorporating group information for analyzing relationships between omics datasets.
• To develop novel group partial least square (gPLS) and sparse group partial least square (sgPLS) algorithms.
• To improve the analysis of associations between different omics data types or omics data and phenotypes.

Main Methods:

• Development of group PLS (gPLS) and sparse group PLS (sgPLS) algorithms.
• Implementation of the gPLS and sgPLS methods in a user-friendly R package named sgPLS.
• Validation of the proposed methods through simulation studies and a real-world HIV therapeutic vaccine trial.

Main Results:

• gPLS and sgPLS demonstrate superior performance compared to sPLS for grouped omics data.
• The developed methods effectively reveal relationships between gene abundance and immunological response in the context of a vaccine trial.
• Parsimonious models are generated, highlighting key associations between biological markers and outcomes.

Conclusions:

• gPLS and sgPLS offer powerful and effective extensions to PLS for omics data integration when biological structures are considered.
• These methods enhance the interpretability and biological relevance of findings from complex omics studies.
• The sgPLS R package provides accessible tools for researchers to apply these advanced analytical techniques.