Visualizing the spatial gene expression organization in the brain through non-linear similarity embeddings

Area of Science:

• Neuroscience
• Bioinformatics
• Computational Biology

Background:

• The Allen Brain Atlases provide genome-wide gene expression data across the mammalian brain.
• Linear dimensionality reduction methods like PCA and cMDS have been used to explore spatial gene expression patterns.
• Understanding the spatial organization of gene expression is crucial for neuroscience research.

Purpose of the Study:

• To introduce and evaluate Barnes-Hut Stochastic Neighbor Embedding (BH-SNE) for analyzing spatially resolved gene expression data.
• To compare the effectiveness of BH-SNE with traditional methods (PCA, cMDS) in mapping brain atlases.
• To assess the ability of BH-SNE to reveal local and global spatial transcriptome structures.

Main Methods:

• Application of Barnes-Hut Stochastic Neighbor Embedding (BH-SNE), a non-linear dimensionality reduction technique.
• Analysis of gene expression data from the Allen Brain Atlases (mouse and human).
• Quantitative comparison of BH-SNE embeddings with Principal Component Analysis (PCA) and classical Multi-Dimensional Scaling (cMDS).

Main Results:

• BH-SNE generated consistent 2D embeddings for mouse and human brain atlases.
• BH-SNE demonstrated superior separation of neuroanatomical regions compared to PCA and cMDS.
• The study assessed the impact of principal components on the global structure of BH-SNE maps.

Conclusions:

• BH-SNE provides comprehensive and intuitive insights into spatial transcriptome structure.
• BH-SNE maps, with or without PCA pre-processing, effectively reveal local and global patterns.
• This non-linear embedding technique enhances the exploration of complex brain gene expression data.