Mixture models with multiple levels, with application to the analysis of multifactor gene expression data.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• High-throughput gene expression studies generate large, high-dimensional datasets.
• Effective data summarization is crucial for biological insight and gene function annotation.
• Existing model-based clustering methods struggle with complex, multi-factor experimental designs.

Purpose of the Study:

• To develop novel model selection procedures for model-based clustering.
• To propose a multilevel mixture model for enhanced interpretability of gene expression data.
• To improve the identification of biological pathways and gene functions.

Main Methods:

• Proposed a multilevel mixture model allowing sparse representations within and between clusters.
• Explored flexible within-cluster parameterizations for enhanced interpretability.
• Implemented sparse between-cluster representations with varying cluster numbers across experimental factors.

Main Results:

• The multilevel mixture model demonstrated improved interpretability in complex, multi-factor contexts.
• Applied the model to a proliferating cell line expression dataset, identifying annotational context and regulatory motifs.
• Evaluated the model's performance on simulated datasets, confirming its effectiveness.

Conclusions:

• The proposed multilevel mixture model offers a powerful tool for summarizing high-dimensional gene expression data.
• Enhanced interpretability of clusters aids in discovering biologically relevant gene groups and pathways.
• This approach advances the analysis of complex biological systems and gene function prediction.