Biologically Informed procedure for Feature Summarization in Spatial Transcriptomics

Modern machine learning approaches have shown remarkable success in extracting patterns from high-dimensional biological data. However, when applied to spatial transcriptomics, these methods face significant challenges due to the sparsity of spatially resolved measurements and the complex, nonlinear relationships between molecular features. To address these challenges, we propose a procedure that integrates single-cell and spatial transcriptomics by considering biologically meaningful regulatory factors as an interpretable feature space. These factors act as latent variables that encode transcriptional programmes, reducing dimensionality and preserving mechanistic relevance. This approach improves interpretability by shifting from raw gene expression to a structured representation of regulatory activity, providing a scalable and biologically interpretable framework for spatial transcriptomic analysis.