Joint modeling of an outcome variable and integrated omics datasets using GLM-PO2PLS

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Disease Modeling

Background:

• Human disease studies often analyze omics datasets individually, overlooking inter-omics relationships.
• Understanding the joint molecular basis of disease requires integrated analysis of multiple omics data types.
• Existing methods include dimension reduction or two-stage approaches, but holistic one-stage models are lacking.

Purpose of the Study:

• To propose a novel one-stage statistical method for jointly modeling multiple omics data and an outcome variable.
• To establish model identifiability and develop algorithms for parameter estimation.
• To derive test statistics for inferring associations between omics and outcomes.

Main Methods:

• Developed a novel one-stage statistical model integrating multiple omics and outcome variables.
• Established model identifiability and employed EM algorithms for maximum likelihood estimation.
• Proposed test statistics and derived their asymptotic distributions for association inference.

Main Results:

• The proposed method allows for joint modeling of omics data and disease outcomes.
• Identifiability and parameter estimation methods were established for normal and Bernoulli outcomes.
• Simulation studies demonstrated the model's effectiveness in evaluating associations.

Conclusions:

• The novel one-stage method provides a holistic approach to analyzing multi-omics data in disease studies.
• Jointly modeling methylation and glycomics with Down syndrome offers enhanced insights compared to individual analyses.
• This integrated approach advances the understanding of complex disease mechanisms.