SPACEL: deep learning-based characterization of spatial transcriptome architectures.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics (ST) technologies map mRNA expression within tissue context.
• Analyzing multiple ST slices and reconstructing 3D tissue architecture are significant challenges.

Purpose of the Study:

• To introduce SPACEL, a deep learning framework for comprehensive ST data analysis.
• To address limitations in joint analysis and 3D reconstruction of ST data.

Main Methods:

• Spoint module: Uses a multilayer perceptron and probabilistic model for cell type deconvolution in ST spots.
• Splane module: Employs graph convolutional networks and adversarial learning for cross-slice spatial domain identification.
• Scube module: Automates coordinate transformation and stacking of ST slices for 3D tissue reconstruction.

Main Results:

• SPACEL demonstrated superior performance in cell type deconvolution compared to 19 existing methods.
• The framework accurately identified spatially coherent domains across multiple ST slices.
• SPACEL achieved precise 3D alignment and reconstruction of tissue architecture from diverse ST datasets.

Conclusions:

• SPACEL provides an integrated and high-performance toolkit for spatial transcriptomics data processing.
• The deep learning approach significantly improves the analysis of multi-slice ST data and 3D tissue reconstruction.
• SPACEL represents a valuable advancement for studying transcriptomic spatial organization in 3D.