PairGP: Gaussian process modeling of longitudinal data from paired multi-condition studies.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput technologies generate complex gene expression time-series data.
• Existing algorithms often overlook the critical pairing effect in longitudinal studies.
• Accurate analysis requires methods that handle non-stationarity and temporal correlations.

Purpose of the Study:

• To introduce PairGP, a non-stationary Gaussian process model for gene expression time-series analysis.
• To account for paired replicates in longitudinal study designs.
• To identify distinct gene expression dynamics across multiple conditions.

Main Methods:

• Developed PairGP, a non-stationary Gaussian process framework.
• Incorporated modeling of paired replicate effects as a distinct component.
• Applied the method to simulated and real RNA sequencing (RNA-seq) time-series data.

Main Results:

• PairGP accurately identifies groups of conditions with differing gene expression dynamics.
• Modeling the pairing effect significantly improves the identification of condition-specific dynamics.
• The method demonstrates robust performance in selecting the most probable groupings, even with numerous conditions.

Conclusions:

• PairGP offers a powerful tool for analyzing gene expression time-series data, particularly in paired designs.
• Explicitly modeling replicate pairing enhances model accuracy and explanatory power.
• The method is broadly applicable to various gene expression time-series datasets.