Denoising image-based spatial transcriptomics data with DenoIST.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Image-based spatial transcriptomics (IST) offers high-resolution gene expression data within tissues.
• Imperfect cell segmentation in IST leads to cross-contamination of gene expression profiles.
• This contamination obscures true biological signals and complicates downstream analysis.

Purpose of the Study:

• To develop a computational tool, DenoIST, for accurate identification and removal of contaminating transcripts in IST data.
• To enhance the specificity and interpretability of gene expression data generated by IST technologies.
• To improve cell type annotation and biological structure resolution in spatial transcriptomics.

Main Methods:

• DenoIST employs a Poisson mixture model to capture local neighborhood contamination.
• The model explicitly accounts for transcript spillover between adjacent cells.
• The tool was validated on multiple real-world IST datasets with varying cell densities.

Main Results:

• DenoIST effectively identifies and removes contaminating transcripts, restoring gene expression specificity.
• The denoised data reveal clearer local biological structures by filtering spurious signals.
• Cell type annotation becomes more consistent and interpretable, reducing ambiguous cell profiles.

Conclusions:

• DenoIST significantly improves the biological interpretability and robustness of IST data.
• The tool can be seamlessly integrated into existing IST analysis workflows.
• By mitigating cross-contamination, DenoIST enhances the reliability of spatial transcriptomics for biological discovery.