Multiset sparse partial least squares path modeling for high dimensional omics data analysis.

Area of Science:

• Computational biology
• Statistical genetics
• Bioinformatics

Background:

• High-dimensional omics data analysis is challenging due to the complexity and volume of data.
• Existing methods struggle to integrate multiple omics data sources and account for their hierarchical structure.
• Interpreting genomewide results from multi-omics analyses remains a significant hurdle.

Purpose of the Study:

• To propose a novel statistical method, multiset sparse Partial Least Squares path modeling (msPLS), for simultaneous analysis of multiple omics data.
• To address the challenges of integrating hierarchical omics data and providing interpretable results.
• To model biological pathways and genetic architectures of complex phenotypes using multi-omics data.

Main Methods:

• Developed msPLS, a generalized penalized form of Partial Least Squares path modeling.
• msPLS simultaneously models effects of molecular markers from multiple omics domains on phenotypic variables.
• The method incorporates relationships between data sources and provides sparse, interpretable results.

Main Results:

• Simulation studies validated msPLS's ability to discover associated variables in high-dimensional data.
• Applied msPLS to omics datasets for Marfan syndrome and Chronic Lymphocytic Leukaemia, exploring associated biological pathways.
• Compared msPLS with Multi-Omics Factor Analysis (MOFA), demonstrating superior performance in certain aspects.

Conclusions:

• msPLS is an effective multivariate method for integrative analysis of multiple high-dimensional omics data.
• The method provides interpretable biomarker identification through sparse solutions.
• msPLS identified relevant biological pathways for Marfan syndrome and CLL, outperforming MOFA in explaining variation in CLL data.