BLOG: Bayesian longitudinal omics with group constraints

Area of Science:

• Biostatistics
• Computational Biology
• Genomics

Background:

• Clinical researchers seek computational biomarkers from short-term longitudinal omics data.
• Accurate biomarker identification is crucial for disease understanding and treatment.

Purpose of the Study:

• To develop and evaluate Bayesian regression and variable selection methods for longitudinal omics data.
• To enhance biomarker discovery and control false discovery rates.

Main Methods:

• Univariate approach using Zellner's g-prior with SURE or sqrt(n) tuning.
• Multivariate approach using Bayesian group lasso with spike and slab priors.
• Utilized first difference (Δ) scale for longitudinal predictors and responses.

Main Results:

• Zellner's g-prior approach demonstrated high specificity and sensitivity in identifying target metabolites.
• Bayesian group lasso also effectively selected target metabolites in simulations.
• Methods were compared against linear mixed effect models on simulated and Tuberculosis study data.

Conclusions:

• The proposed Bayesian methods offer improved inference and prediction for biomarker identification.
• Automated hyperparameter selection enhances the robustness of the Zellner's g-prior approach.
• These computational tools advance the discovery of reliable biomarkers from complex omics datasets.