DiffuST: A Latent Diffusion Model for Spatial Transcriptomics Denoising

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics technologies offer gene expression profiling with spatial context.
• Noise from low RNA capture and experimental steps can obscure biological signals.
• Existing denoising methods may not fully address spatial transcriptomics data complexities.

Purpose of the Study:

• To develop an advanced computational method for denoising spatial transcriptomics data.
• To improve the accuracy and interpretability of spatial transcriptomics analyses.
• To leverage pathology images alongside gene expression data for enhanced denoising.

Main Methods:

• Development of DiffuST, a latent diffusion model tailored for spatial transcriptomics.
• Integration of a graph autoencoder for multi-scale feature extraction.
• Utilizing pre-trained models and pathology images to guide the denoising process.
• Mapping multi-scale features into a shared latent space for effective noise reduction.

Main Results:

• DiffuST demonstrated superior performance compared to existing denoising methods across multiple datasets.
• The method effectively reduced noise while preserving crucial biological signals.
• Denoised data from DiffuST led to enhanced downstream spatial transcriptomics analyses.
• Significant biological insights were uncovered using the denoised data.

Conclusions:

• DiffuST provides a powerful new tool for denoising spatial transcriptomics data.
• The method improves the reliability and biological interpretability of spatial transcriptomics studies.
• DiffuST has the potential to advance discoveries in various biological fields utilizing spatial omics.