Dynamic factor analysis with dependent Gaussian processes for high-dimensional gene expression trajectories

Area of Science:

• Genomics
• Systems Biology
• Computational Biology

Background:

• High-dimensional, longitudinal gene expression data are crucial for understanding biological mechanisms in precision medicine.
• Complex diseases may be best understood by analyzing interacting biological pathways rather than individual genes.

Purpose of the Study:

• To develop a Bayesian approach for characterizing correlations among biological pathways using longitudinal gene expression data.
• To map high-dimensional gene expression trajectories to low-dimensional pathway trajectories, relaxing the assumption of independent factors.

Main Methods:

• Utilized dependent Gaussian processes (DGP) to model pathway correlations.
• Employed Bayesian sparse factor analysis to map gene expression to pathway trajectories.
• Developed a Monte Carlo expectation maximization (MCEM) scheme for model fitting, integrated with Markov Chain Monte Carlo (MCMC) and an R package (GPFDA).

Main Results:

• The proposed method demonstrated superior performance in recovering pathway expression trajectories.
• Successfully revealed relationships between genes and pathways.
• Achieved improved gene expression prediction with closer point estimates and narrower predictive intervals compared to existing methods.
• Validated through simulations and real data analysis.

Conclusions:

• The novel Bayesian approach effectively models correlated biological pathways from longitudinal gene expression data.
• The method enhances understanding of gene-pathway relationships and improves predictive accuracy for precision medicine applications.
• The associated R package (DGP4LCF) is publicly available, facilitating broader adoption and further research.