A probabilistic latent semantic analysis model for coclustering the mouse brain atlas

Area of Science:

• Neuroscience
• Computational Biology
• Genomics

Background:

• Mammalian brain development involves complex gene expression patterns.
• Understanding these patterns is crucial for characterizing cell types and functions.
• Existing methods may not fully capture the spatiotemporal dynamics of gene expression.

Purpose of the Study:

• To develop and evaluate a novel coclustering method for analyzing spatiotemporal gene expression data.
• To apply this method to the Allen Developing Mouse Brain Atlas.
• To identify co-expressed genes and improve voxel clustering consistency with neuroanatomy.

Main Methods:

• Employed a graph approximation formulation for simultaneous coclustering of genes and brain voxels.
• Expressed the formulation as a probabilistic latent semantic analysis (PLSA) model.
• Utilized the expectation-maximization algorithm for PLSA parameter estimation and evaluated on synthetic data.

Main Results:

• The proposed coclustering method outperformed prior methods on synthetic datasets.
• Voxel clustering derived from the Allen Developing Mouse Brain Atlas showed improved consistency with classical neuroanatomy.
• Identified sets of genes exhibiting co-expression within specific brain regions.

Conclusions:

• The developed coclustering approach offers a robust framework for analyzing complex brain gene expression data.
• This method enhances the interpretation of spatiotemporal gene regulation during brain development.
• The findings contribute to a deeper understanding of brain organization and cellular heterogeneity.