Enhancing Spatial Transcriptomics via Spatially Constrained Matrix Decomposition with EDGES.

Area of Science:

• Biomedical research
• Genomics
• Computational biology

Background:

• Spatial transcriptomics (ST) offers tissue architecture insights but faces gene detection and accuracy limitations.
• Current computational methods struggle with spatial context integration and biases from single-cell RNA sequencing data.

Purpose of the Study:

• To develop a novel computational framework, EDGES, to address limitations in imaging-based ST data.
• To simultaneously predict undetected gene expression and denoise transcriptional profiles.

Main Methods:

• Developed EDGES, a spatially constrained non-negative matrix factorization framework.
• Incorporated spatial information using graph Laplacian regularization.
• Integrated cellular and gene-specific representations for accurate predictions.

Main Results:

• EDGES demonstrated superior predictive performance in comprehensive evaluations.
• The framework outperformed existing denoising methods for ST data.
• Identified novel biological markers and spatially resolved expression patterns.

Conclusions:

• EDGES enhances the reliability of imaging-based ST data.
• Facilitates more accurate and biologically meaningful interpretations of ST data.
• Provides an advanced tool for biomedical research discoveries.