Unsupervised pattern identification in spatial gene expression atlas reveals mouse brain regions beyond established...

Area of Science:

• Neuroscience
• Computational Biology
• Genomics

Background:

• Large-scale spatial gene expression data is rapidly growing.
• Efficient computational tools are needed to analyze this data in its spatial context.
• Understanding gene expression trends and anatomical localization is crucial for brain research.

Purpose of the Study:

• To develop and validate a novel computational approach for analyzing 3D spatial gene expression data.
• To identify principal patterns (PPs) of gene expression in the whole mouse brain.
• To correlate these PPs with anatomical regions and build a gene expression-based brain ontology.

Main Methods:

• Applied stability-driven unsupervised learning (staNMF) to whole mouse brain spatial gene expression data.
• Performed spatial correlation analysis comparing identified PPs to the Allen Mouse Brain Atlas.
• Evaluated the performance of staNMF against principal component analysis and clustering algorithms.

Main Results:

• Identified stable and spatially coherent PPs that accurately approximate spatial gene data.
• Demonstrated high correlation between PPs and combinations of expert-annotated brain regions.
• Developed a novel brain ontology derived solely from spatial gene expression data.
• Showcased staNMF's superior performance compared to traditional methods.
• Revealed regional imbalances in brain-wide genetic architecture and identified region-specific genes and coexpression networks.

Conclusions:

• Stability-driven machine learning, specifically staNMF, is advantageous for biological discovery from dense spatial gene expression data.
• The identified PPs provide a new framework for understanding brain organization and function.
• This approach streamlines complex analyses that are challenging for manual methods.