Generative AI Enhances Spatial Gene Expression Analysis in Tissue Samples

Spatial transcriptomics is a cutting-edge technique that examines gene expression within tissue sections, such as the heart, skin, or liver. By mapping gene activity in its spatial context, researchers gain valuable insights into how cellular organization influences biological processes and diseases.

Until now, scientists using spatial transcriptomics faced a trade-off: they could either achieve genome-wide coverage or single-cell resolution—but not both.

To overcome this limitation, researchers from St. Jude Children’s Research Hospital and the University of Wisconsin-Madison have developed a computational tool that leverages generative artificial intelligence (AI) to enhance the resolution of sequencing-based spatial transcriptomics while preserving comprehensive gene coverage.

Their algorithm and initial findings were recently published in Nature Methods.

"We have developed the first generative algorithm capable of predicting spatial gene expression at the single-cell level across the entire transcriptome. The key innovation is integrating data from single-cell RNA sequencing atlases and histology imaging using a generative modeling approach, which enables full transcriptome coverage with single-cell resolution in spatial RNA sequencing."

Jiyang Yu, PhD, Study Co-Senior and Corresponding Author, Interim Chair, Department of Computational Biology, St. Jude Children’s Research Hospital

The newly developed computational tool, Spot Imager with Pseudo Single-Cell Resolution Histology (Spotiphy), employs a machine learning algorithm to significantly enhance conventional spatial transcriptomics techniques.

Traditional methods rely on analyzing predefined “spots” on a grid to capture gene expression, where each spot represents a cluster of multiple, often heterogeneous, cells. This aggregation makes it challenging to classify and analyze gene activity at the single-cell level.

Spotiphy addresses this by applying machine learning to extrapolate cell-type proportions and gene expression data, effectively filling in the gaps between these “spots.” The algorithm is trained on vast spatial transcriptomics datasets paired with histological images, allowing it to accurately predict missing details.

"Imagine looking at a picture of a hand where the middle section is missing. The algorithm has learned general rules from its training data, enabling it to reconstruct the missing part—just as it fills in the spaces between imaging spots in spatial transcriptomics."

Junmin Peng, PhD, Study Co-Senior Author, Departments of Structural Biology and Developmental Neurobiology, St. Jude Children’s Research Hospital

Peng, whose research focuses on neurobiology, emphasized how single-cell spatial transcriptomics can improve the study of neurodegenerative diseases like Alzheimer’s by providing a clearer view of aberrant cells and their gene expression patterns.

"Previous spatial transcriptomics approaches couldn’t truly resolve single-cell data. Existing methods produce low-resolution outputs, often clustering multiple cells within a single spot. By merging adjacent tissue sections for one-to-one RNA analysis and imaging, Spotiphy achieves clean single-cell resolution with high gene coverage."

Junmin Peng, PhD

When applied to a mouse model of Alzheimer’s disease, Spotiphy validated previous research by accurately identifying cell locations and gene expression patterns, reinforcing confidence in the technology.

In new experiments, researchers used Spotiphy to uncover previously undetectable cellular distinctions. For instance, they found that specific subsets of astrocytes, a major cell type in the central nervous system, are localized to certain brain regions. They also observed an increased presence of disease-associated microglia—a rare immune cell type—in Alzheimer’s-affected brains, supporting existing theories that microglial dysfunction plays a role in the disease.

"The real power of this algorithm lies in its ability to identify subtle differences within the same cell type—distinctions that previous technologies couldn’t detect. For example, it allows us to pinpoint subpopulations of astrocytes in specific brain regions."

Jiyang Yu, PhD

Spotiphy’s capabilities extend beyond neurodegenerative diseases. The research team demonstrated its effectiveness in analyzing other tissues, including cancer samples. The tool successfully identified distinct spatial domains and changes in tumor-tumor microenvironment interactions.

"We observed that our data reflected known breast cancer heterogeneity. Additionally, we invested significant effort in generating matched datasets for mouse brains, which we believe will be a valuable resource for the spatial omics community."

Jiyuan Yang, PhD, Co-First Author, St. Jude Department of Computational Biology

Yu concluded, "With Spotiphy, we have developed a tool that enables single-cell spatial transcriptomics for any given tissue. In other words, it allows scientists to see what was previously invisible."