Clustering of time-course gene expression profiles using normal mixture models with autoregressive random effects

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Traditional Fourier series approximations inadequately model complex time-course gene expression data.
• Existing methods often overlook temporal dependencies and correlations among gene profiles.
• Periodic gene expression, like in yeast cell cycle data, requires advanced modeling techniques.

Purpose of the Study:

• To propose a novel mixture model for clustering time-course gene expression profiles.
• To address limitations of existing methods in capturing data complexity and inter-gene correlations.
• To enhance the reliability and robustness of gene expression data clustering.

Main Methods:

• Development of a mixture model incorporating autoregressive random effects of the first order.
• Extension of the EMMIX-WIRE procedure for enhanced clustering.
• Utilizing gene-specific random effects with autocorrelation variance structure.
• Implementation in a flexible R package for user-defined random effects.

Main Results:

• The proposed model demonstrates superior performance over existing methods for time-course data clustering.
• Achieves more reliable and robust clustering, particularly when gene profiles are correlated.
• Provides comparable results even with weak correlations between gene profiles.
• Identifies relevant clusters of coregulated genes in real datasets, validated by functional annotation.

Conclusions:

• The new model offers a more reliable and robust approach to clustering time-course data.
• It effectively models correlations among observations and gene coregulation through random effects.
• The associated R package provides flexibility for improved modeling and clustering outcomes.