DiffGSP: reversing mRNA diffusion to unlock high-fidelity spatial transcriptomics

Spatial transcriptomics has revolutionized our understanding of gene expression within tissues, but a major hurdle remains: mRNA diffusion during sample preparation can blur spatial relationships between cells. This spatial misalignment undermines the accuracy of spatial gene expression profiles, leading to potential misinterpretation of biological processes.

Enter DiffGSP, a novel physics-informed framework designed to tackle this issue. By combining Fick's law of diffusion with graph signal processing, DiffGSP can model and correct the distortions introduced by mRNA migration. This innovative approach restores the true spatial arrangement of gene expression patterns, offering a more faithful representation of cellular organization.

DiffGSP has been rigorously tested across a wide range of datasets, consistently demonstrating its ability to recover accurate spatial gene expression profiles. The results are nothing short of impressive: in the mouse brain, DiffGSP uncovers fine anatomical structures; in the kidney, it reveals spatially organized gene modules; in colorectal cancer, it highlights intratumoral heterogeneity; and in lung adenocarcinoma, it brings out tertiary lymphoid structures.

What sets DiffGSP apart is its foundation on a physically interpretable framework, making it not just a computational tool, but a conceptually sound one. Moreover, its applicability extends to various sequencing-based spatial transcriptomics technologies, promising a broad impact on the field.

By effectively reversing the effects of mRNA diffusion, DiffGSP elevates the fidelity of spatial gene expression reconstruction. This advancement not only enhances the reliability of spatial transcriptomics, but also opens up new avenues for more reliable biological interpretation, paving the way for deeper insights into the complex architecture of tissues.