SpatialDDLS

SpatialDDLS applies neural network-based deconvolution to spatial transcriptomics data to infer spot-level cell-type compositions using single-cell RNA sequencing (scRNA-seq)-derived simulated mixtures.

Key Features:

• scRNA-seq-based simulation: Simulates mixed transcriptional profiles with predefined cellular compositions using single-cell RNA sequencing data.
• Fully connected neural network: Trains a fully connected neural network on simulated mixtures to learn mappings from mixed profiles to cellular compositions.
• Spot-level deconvolution: Deconvolves cell-type compositions within individual spatial transcriptomics spots.
• Predefined cellular compositions: Uses predefined cell-type proportions in simulations to generate training data for the model.
• Performance: Employs a fast and efficient algorithm that is reported to outperform existing state-of-the-art methods in accuracy and speed.
• Cell-type diversity inference: Predicts the diversity and relative abundance of cell types at each spatial spot.

Scientific Applications:

• Tissue structure and cellular organization: Infers spatial distributions of cell types to analyze tissue architecture and cellular arrangement.
• Developmental biology: Resolves spatially patterned cell-type compositions relevant to developmental processes.
• Oncology: Maps tumor and microenvironment cell-type distributions within spatial transcriptomics profiles.
• Immunology: Characterizes spatial distributions of immune cell types in tissue contexts.

Methodology:

Uses scRNA-seq to simulate mixed transcriptional profiles with predefined cellular compositions, trains a fully connected neural network on those simulations, and applies the trained network to deconvolve cell-type compositions of spatial transcriptomics spots.