S2potAE: multimodal spatial spot autoencoder integrating image and transcriptomic features for deconvolution.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics (ST) offers insights into tissue architecture and cellular heterogeneity.
• Accurate cell-type deconvolution in ST data is hindered by resolution differences and batch effects.

Purpose of the Study:

• Introduce S$^{2}$potAE, a novel spatial spot autoencoder framework for precise spot-level deconvolution.
• Integrate gene expression, spatial coordinates, and histological morphology for enhanced deconvolution.
• Improve biological relevance and interpretability through an auxiliary pathological classification task.

Main Methods:

• Developed S$^{2}$potAE, a graph-based spatial encoder and image embedding framework.
• Employed multilevel feature aggregation to fuse spatially-aware features.
• Validated using simulated and real datasets from human and mouse tissues.

Main Results:

• S$^{2}$potAE demonstrated superior accuracy, robustness, and interpretability over existing methods.
• The framework effectively resolved complex cellular compositions and identified tumor boundaries.
• Achieved nuanced cell-type distribution mapping in diverse biological samples.

Conclusions:

• S$^{2}$potAE significantly advances spatial transcriptomics deconvolution.
• The approach enhances the utility of ST in biological research and clinical applications.
• Provides a robust tool for analyzing cellular heterogeneity in complex tissues.