BANKSY

BANKSY integrates cell-type clustering and tissue-domain segmentation for spatial omics data by embedding cells into a product space that combines intrinsic transcriptomic profiles with local microenvironment features to analyze cellular states and tissue organization.

Key Features:

• Unified Framework: Merges cell type clustering and tissue domain segmentation into a single analytical framework.
• Spatial Feature Augmentation: Augments cell-level data with local microenvironment information using a neighborhood kernel to capture spatial context.
• Product-space Embedding: Embeds cells into a product space that combines intrinsic transcriptomic profiles with neighborhood-derived features.
• Versatility Across Data Types: Validated across RNA imaging, sequencing, and protein imaging modalities.
• Biological Insights: Enables identification of niche-dependent cellular states, including observations in the mouse brain.
• Benchmark Performance: Demonstrated superior performance in comparative analyses for domain segmentation and cell typing tasks.
• Quality Control and Batch Effect Correction: Supports quality control of spatial transcriptomics data and spatially aware batch effect correction.
• Scalability and Efficiency: Scales to datasets with millions of cells with greater processing speed and scalability than traditional methods.

Scientific Applications:

• Spatial omics analysis: Integration and analysis of spatially resolved omics datasets, including RNA imaging, sequencing, and protein imaging, for joint cell-state and domain mapping.
• Developmental biology: Studying spatial organization and cell-state dynamics during development.
• Neuroscience: Mapping tissue domains and niche-dependent cellular states in complex tissues such as the mouse brain.
• Oncology: Analyzing spatial organization and cell-type distributions in cancer tissues.
• Spatial transcriptomics quality control: Providing dataset-level quality control and spatially informed batch correction for large-scale studies.

Methodology:

Embeds cells into a product space combining intrinsic transcriptomic profiles with local microenvironment features using a neighborhood kernel and jointly performs cell-type clustering and tissue-domain segmentation.