SPROUT

SPROUT reconstructs single-cell-resolution spatial structures from transcriptomics data to infer cell proximities and ligand-receptor-mediated interactions, bridging single-cell RNA sequencing and spatial transcriptomics.

Key Features:

• Ligand-Receptor Interaction Analysis: Leverages ligand-receptor interactions to infer spatially constrained cell proximities and identify dominant ligand-receptor pairs between neighboring cells at single-cell resolution.
• Pseudo Affinity Reduction: Reduces pseudo affinities in intercellular affinity matrices using partial correlation and spectral graph sparsification to retain genuine interactions.
• Low-Dimensional Embedding: Embeds estimated interactions into a low-dimensional space by optimizing a cross-entropy objective to restore intercellular structure.
• Representative Single-Cell Profile Curation: Curates representative single-cell profiles for each spatial spot from a candidate library to link scRNA-seq profiles to spatial locations.
• Spatial Coordinates Refinement: Refines spatial coordinates based on the refined affinity estimates and embeddings.
• Performance Metrics: Demonstrated shape Pearson correlations of 0.91–0.97 on datasets such as mouse hippocampus and human organ tumor microenvironments.
• De Novo Reconstruction Capability: Performs de novo reconstruction without prior spatial information, achieving cell-type proximity correlations of 0.68 and 0.89 compared to immunohistochemistry-informed structures in human developing heart and tumor microenvironment datasets.

Scientific Applications:

• Tissue Architecture Reconstruction: Reconstructs spatial organization of tissues such as mouse hippocampus and human organ tumor microenvironments from transcriptomics data.
• Cell–Cell Interaction Inference: Identifies and ranks dominant ligand-receptor pairs between neighboring cells at single-cell resolution.
• Microenvironment and Developmental Analysis: Analyzes tumor microenvironments and developing organs (e.g., human developing heart) and compares reconstructed structures to immunohistochemistry-informed spatial maps.

Methodology:

Curates representative single-cell profiles for each spatial spot from a candidate library; reduces pseudo affinities in intercellular affinity matrices using partial correlation and spectral graph sparsification; embeds estimated interactions into a low-dimensional space optimizing a cross-entropy objective; and refines spatial coordinates.