STEM enables mapping of single-cell and spatial transcriptomics data with transfer learning

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics (ST) lacks single-cell resolution, while single-cell RNA sequencing (SC) lacks spatial information.
• Integrating ST and SC data is crucial for understanding tissue physiology and pathology at a granular level.

Purpose of the Study:

• To develop a computational method for integrating ST and SC data.
• To enable the creation of single-cell resolution spatial transcriptomic landscapes.
• To uncover the spatial organization and heterogeneity of cells within tissues.

Main Methods:

• Developed STEM (SpaTially aware EMbedding), a deep transfer learning method.
• Encoded ST and SC data into a unified, spatially aware embedding space.
• Inferred SC-ST mapping and predicted pseudo-spatial adjacency for SC data.

Main Results:

• STEM effectively integrated ST and SC data, preserving spatial information.
• The method eliminated technical biases between the two data types.
• Applied to human cancer and liver datasets, STEM identified rare cell localizations and spatial gene expression variations.

Conclusions:

• STEM enables the construction of single-cell level spatial transcriptomic maps by integrating ST and SC data.
• This approach provides mechanistic insights into tissue microenvironments and spatial heterogeneity.
• STEM is a powerful tool for advancing spatial biology research.