SpaDiff denoises sequence-based spatial transcriptomics via diffusion process

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Sequence-based spatial transcriptomics enables gene expression mapping within tissues.
• Spot-swapping is an artifact that redistributes RNA molecules, distorting spatial data.
• This distortion biases downstream analyses, hindering accurate biological interpretation.

Purpose of the Study:

• To introduce SpaDiff, a novel denoising framework for spatial transcriptomics.
• To correct RNA redistribution artifacts caused by spot-swapping.
• To improve the accuracy of spatial gene expression data and downstream analyses.

Main Methods:

• SpaDiff utilizes a mass-conserving denoising framework.
• The method employs score-guided reverse diffusion to correct artifacts.
• Validation performed on simulated and real spatial transcriptomics datasets.

Main Results:

• SpaDiff consistently improves the recovery of spatial gene expression distributions.
• The framework robustly restores gene-level spatial patterns in simulations.
• SpaDiff accurately relocalizes transcripts in chimeric samples and enhances domain identification in real tissues.

Conclusions:

• SpaDiff effectively corrects spot-swapping artifacts in spatial transcriptomics.
• The framework significantly improves spatial accuracy of gene expression data.
• SpaDiff enables more reliable downstream analyses, including clustering and spatial domain identification.