SPARK

SPARK identifies genes with spatial expression patterns from spatially resolved transcriptomic datasets such as Spatial Transcriptomics, Slide-seq, seqFISH, and MERFISH to characterize spatial heterogeneity across tissue sections.

Key Features:

• Statistical modeling: Employs generalized linear spatial models and statistical tests formulated to directly model spatial count data for detection of spatial expression patterns.
• Control of type I errors: Implements procedures to control type I error rates and reduce false positives in spatial gene identification.
• High statistical power: Demonstrates substantially higher statistical power—reported up to ten times more powerful—relative to existing methods.
• Scalability and computational efficiency: Utilizes a penalized quasi-likelihood approach to scale to tens of thousands of genes across similarly large numbers of samples while maintaining computational efficiency.

Scientific Applications:

• Spatial transcriptomic landscape characterization: Identify spatially variable genes to map gene expression variation across tissue regions.
• Developmental biology: Resolve spatial gene expression patterns relevant to tissue development and morphogenesis.
• Oncology: Detect spatial heterogeneity of gene expression within tumors to inform cancer biology studies.
• Analysis of published spatial datasets: Apply to Spatial Transcriptomics, Slide-seq, seqFISH, and MERFISH datasets to reveal novel spatial expression patterns.

Methodology:

Fits generalized linear spatial models to spatial count data, applies statistical tests formulated for hypothesis testing with control of type I error, and implements a penalized quasi-likelihood algorithm for computational efficiency and scalability.