MULTI-WAY BLOCKMODELS FOR ANALYZING COORDINATED HIGH-DIMENSIONAL RESPONSES.

Area of Science:

• Computational Biology
• Systems Biology
• Statistical Modeling

Background:

• Quantifying temporal coordination in high-dimensional biological responses is challenging.
• Existing methods often rely on pre-processing steps like binning and thresholding, which can introduce bias.

Purpose of the Study:

• To introduce a novel family of multi-way stochastic blockmodels for quantifying temporal coordination.
• To develop and evaluate inference procedures for these models.
• To analyze the functional mechanisms driving coordination between gene expression and metabolite concentrations.

Main Methods:

• Development of multi-way stochastic blockmodels.
• Inference procedures using collapsed Gibbs sampling and variational methods.
• Evaluation on simulated data and empirical analysis in *S. cerevisiae*.

Main Results:

• The proposed models effectively quantify temporal coordination without pre-processing.
• Comparison of inference methods (Gibbs sampling vs. variational) in terms of estimation accuracy and runtime.
• Quantification of the impact of binning and thresholding on estimation tasks.

Conclusions:

• Multi-way stochastic blockmodels provide a robust framework for analyzing temporal coordination in high-dimensional biological data.
• The developed inference methods are efficient and accurate.
• The study offers insights into the coordination of gene expression and metabolite concentrations during nutrient starvation in *S. cerevisiae*.