3D reconstruction of spatial transcriptomics with spatial pattern enhanced graph convolutional neural network.

Area of Science:

• Biomedical research
• Computational biology
• Genomics

Background:

• Spatially resolved transcriptomics (SRT) offers gene expression and spatial data but current analysis methods are limited to 2D.
• Existing 2D approaches cannot fully capture the complexity of tissue structure, cell communication, and developmental trajectories in three dimensions.

Purpose of the Study:

• To develop a novel computational framework, Spa3D, for reconstructing and analyzing 3D spatial structures from 2D SRT data.
• To overcome the limitations of 2D analysis by incorporating physical z-axis information for more accurate biological insights.

Main Methods:

• Spa3D utilizes anti-leakage Fourier transform and graph convolutional neural networks to reconstruct 3D spatial structures from multiple 2D SRT slices.
• The method incorporates physical z-axis distances, enabling robust 3D modeling even with variations between adjacent tissue slices.

Main Results:

• Spa3D accurately identifies spatial domains, elucidates 3D cell-cell communication networks, and models organ-level tempo-spatial development patterns.
• The framework enhances spatial domain detection and reveals 3D spatial trajectories previously undetectable with 2D methods.
• Spa3D demonstrates applicability across various SRT platforms, outperforming existing state-of-the-art methods.

Conclusions:

• Spa3D provides a robust solution for 3D spatial transcriptomics analysis, enabling deeper understanding of tissue complexity.
• This approach facilitates novel biological discoveries by revealing spatial features and developmental patterns in a true 3D context.