STORM: spatial transcriptomics optimization by resolution via matrix factorization

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics (ST) loses critical tissue architecture and spatial information.
• Current ST platforms yield incomplete and noisy data, obscuring biological patterns.
• Technical limitations and tissue variability hinder accurate spatial transcriptomic analysis.

Purpose of the Study:

• Introduce STORM (spatial transcriptomics optimization by resolution via matrix factorization), a machine learning framework.
• Improve the fidelity of spatial transcriptomics data, especially under sparse conditions.
• Enable accurate reconstruction of spatial gene-expression patterns and preserve biological structure.

Main Methods:

• Formulate spatial transcriptomics recovery as a low-rank tensor decomposition problem.
• Integrate multimodal biological priors using a regularization strategy.
• Incorporate spatial continuity, tissue morphology from histology images, and gene-gene interactions from protein-protein interaction networks.

Main Results:

• STORM accurately reconstructs gene expression at unobserved locations.
• The method preserves biologically meaningful spatial structures.
• STORM outperforms state-of-the-art methods in recovering spatial gene-expression patterns in lung tissues.
• The framework remains robust even with a majority of missing spatial measurements.

Conclusions:

• STORM provides a reliable foundation for high-resolution spatial transcriptomic analysis.
• The framework is effective in settings with sparse or incomplete experimental data.
• Explicitly embedding biological structure enhances the reconstruction process for spatial transcriptomics.