Hierarchical Bayesian modelling of gene expression time series across irregularly sampled replicates and clusters

Area of Science:

• Computational Biology
• Statistical Modeling
• Genomics

Background:

• Analyzing time-course gene expression data from microarrays and sequencing requires efficient statistical models.
• Existing methods struggle with temporal dynamics, replication, and irregular sampling, limiting biological signal extraction.
• Tasks like data fusion and clustering are hindered by current limitations.

Purpose of the Study:

• To introduce hierarchical Gaussian processes (HGPs) as a versatile statistical model for gene expression time-series.
• To demonstrate HGP's utility in missing data imputation, data fusion, and clustering.
• To overcome limitations of existing methods regarding data regularity and replication.

Main Methods:

• Developed a hierarchical Gaussian process model tailored for gene expression time-series.
• Applied the model to tasks including missing data imputation, data fusion, and clustering.
• Validated performance on real-world datasets, including those with irregular sampling and replications.

Main Results:

• HGPs significantly outperform common imputation methods for both random and systematic missing data.
• The model produces more biologically relevant clusters by effectively modeling inter- and intra-cluster variance.
• Demonstrated successful application to a Drosophila developmental dataset with irregular sampling.

Conclusions:

• Hierarchical Gaussian processes provide a robust statistical framework for gene expression time-series analysis.
• The HGP model is computationally efficient, requires minimal additional parameters, and is easily integrated into existing algorithms.
• Python implementations are available, facilitating adoption and further research.