spage2vec

spage2vec maps local spatial gene expression patterns into lower-dimensional embeddings to characterize spatial transcriptomic heterogeneity at subcellular resolution using multiplexed in situ RNA detection data without requiring cell segmentation.

Key Features:

• Graph-Based Representation: Models the spatial transcriptomic landscape as a graph where nodes represent local spatial gene expression patterns.
• Machine Learning Embeddings: Employs machine learning to generate lower-dimensional embeddings that preserve biologically meaningful local spatial relationships.
• Unsupervised, Segmentation-Free Analysis: Operates without prior segmentation or complementary single-cell sequencing information.
• Multiplexed in situ RNA Support: Directly analyzes multiplexed in situ RNA detection and in situ transcriptomic data.
• Subcellular Resolution and Scalability: Encodes subcellular-resolution patterns and scales to datasets comprising hundreds of individual cells.

Scientific Applications:

• Mouse brain spatial transcriptomics: Applied to multiplexed spatial gene expression datasets from mouse brain to encode localized gene expression signatures.
• Identification of localized gene expression signatures: Reveals re-occurring localized gene expression signatures linked to cellular and subcellular processes.
• High-resolution and large-scale studies: Supports high-resolution transcriptomic analyses across hundreds of cells to investigate tissue architecture.
• Biological research areas: Applicable to studies of tissue architecture, disease mechanisms, and developmental biology where spatial gene expression is critical.

Methodology:

Constructs a spatial gene expression network from in situ transcriptomic data, trains an unsupervised graph representation model to produce node embeddings, and analyzes those embeddings to identify recurring localized gene expression signatures.