SPACEL

SPACEL characterizes spatial transcriptomics (mRNA) data to deconvolute spot-level cell-type composition, identify spatially coherent domains, and align multiple slices into three-dimensional tissue architectures.

Key Features:

• Spoint Module: Uses a multiple-layer perceptron integrated with a probabilistic model to deconvolute cell type composition within each spot on a single ST slice.
• Splane Module: Applies a graph convolutional network combined with adversarial learning to identify spatial domains that are transcriptomically and spatially coherent across multiple ST slices.
• Scube Module: Automates transformation of spatial coordinate systems between consecutive slices and stacks transformed slices to construct a 3D tissue architecture.

Scientific Applications:

• Cell type deconvolution: Recover and quantify cell-type mixtures at spot-level resolution from spatial transcriptomics data.
• Spatial domain identification: Detect transcriptomically and spatially coherent tissue domains within and across ST slices.
• 3D tissue reconstruction and alignment: Align and stack consecutive ST slices to reconstruct three-dimensional tissue organization from spatial coordinates.

Methodology:

The approach comprises three modules—Spoint (multiple-layer perceptron + probabilistic model), Splane (graph convolutional network + adversarial learning), and Scube (spatial coordinate transformation and stacking); performance was assessed by comparative analyses on simulated and real ST datasets from various tissues and technologies.