SMURF: soft-segmentation for single-cell reconstruction and topological analysis of spatial transcriptomic data.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Accurate transcript assignment to single cells is crucial for high-resolution spatial transcriptomics.
• Existing computational methods face challenges in complex tissue architectures.

Purpose of the Study:

• To introduce SMURF (Segmentation and Manifold UnRolling Framework), a novel deep learning algorithm for soft-segmentation in spatial transcriptomics.
• To enhance the accuracy of mapping mRNAs to individual cells and analyzing tissue organization.

Main Methods:

• SMURF employs deep learning to map mRNAs from capture spots to nearby nuclei.
• The algorithm unrolls complex tissue architectures by projecting cells onto Cartesian coordinates.
• Applied to Visium HD data for segmenting over 400,000 cells in the mouse ileum.

Main Results:

• SMURF demonstrates superior accuracy in assigning mRNAs to single cells compared to existing methods.
• Successfully unrolled complex tissues, revealing zonated transcriptional programs and cell-type organization.
• Identified zonated gene expression programs and regulatory transcription factors in the maturing intestinal villus.
• Revealed gene expression gradients along the proximal-distal axis accumulating in the upper villus.
• Showcased reprogramming of upper villus gene expression by environmental lumenal signals.

Conclusions:

• SMURF is a powerful, cross-platform framework for analyzing gene expression within native tissue contexts.
• The study establishes environmental inputs as key determinants of regional transcriptional identity in the intestine.
• SMURF enables detailed analysis of cell-type organization and gene expression gradients in complex tissues.