New AI system maps gene expression across the whole body

Understanding gene expression within the body has been a boon for 21st century biology and therapeutics, but most discoveries that use these technologies only focus on one organ or one small area of tissue. 

At the University of Chicago's Pritzker School of Molecular Engineering (UChicago PME), Assoc. Prof. Nicolas Chevrier's group has developed a new system to understand how diseases affect molecules, cells, tissues, and organs across the whole body-a major goal of both scientists and physicians. The interdisciplinary work was led by Maggie Clevenger, a staff scientist in the lab, and involved several industrial and academic collaborators.

By creating a new technique to prepare specimens for examination and combining it with computational tools-including a machine learning model-Chevrier and his lab mapped gene expression across whole mouse body sections.

The system accurately mapped all organs, tissue regions, and about 75% of all known cell types in the mouse body, providing a toolkit that researchers can use to study molecular and cellular processes across the entire body of the laboratory mouse. The results, published today in Cell, could be used in both basic science research and in areas such as drug discovery. 

We now have a tool to generate datasets at a scale that was previously unimaginable. It lays a foundation to generate the kind of data which you'd need to build a 'virtual mouse' that could be used to test therapies and understand body-wide biological processes. That's the ultimate goal."

Assoc. Prof. Nicolas Chevrier, University of Chicago's Pritzker School of Molecular Engineering

Measuring systemic inflammation across a whole body

The new technique harnesses spatial transcriptomics, which uses high-resolution microscopy and genetic sequencing to measure gene expression across tissue. This technique, optimized within the last decade, gives researchers important insight into structure and disease within an organ or tissue sample rather than in just single cells. 

But researchers have been hamstrung by the small scales that this technique allows for. Chevrier wanted to use it to measure gene expression across an entire mouse model.

In 2025, he and his team developed Array-seq, which uses DNA microarrays with custom-designed probes to analyze tissue samples.

But to use that Array-seq technology to analyze a whole mouse, they needed to develop methods to generate very thin slices of a frozen mouse body, then transfer it onto Array-seq slides while keeping it intact and preserving the RNA. Working with Prof. Tadafumi Kawamoto of Tsurumi University (Yokohama, Japan), they did just that, getting a cross-section of the whole body of a laboratory mouse that was the thickness of an average cell.

After performing spatial transcriptomics on the specimen, the team then developed a new computational model to annotate the cellular information of the entire mouse. The model was developed in collaboration with a long-term industrial partner of the lab, Ashwini Patil of Combinatics (Chiba, Japan).

The research team also teamed up with AI expert Prof. Feng Bao of Fudan University (Shanghai, China) to create a new machine learning model that labels each organ, tissue and cell type onto tissue sections simply stained with hematoxylin and eosin – the most widely used stain in tissue research and clinical diagnostics. 

"If you were to do this manually, you would need to label all these different cell types with staining reagents such as antibodies in the lab, and it is currently infeasible to do across the mouse body," Chevrier said. "We trained an AI model to do this, so now we can do it virtually and very cheaply."

To test their new technologies, they used them to measure inflammation in a mouse model of sepsis-an organism-wide, dysregulated immune response to infection and a major public health challenge. 

"For the first time, we could quantify the impact of systemic inflammation on every cell type and on every major organ tissue at a scale that was not possible before," Chevrier said. "It paves the way for the molecular mapping of the laboratory mouse and many other model systems at unprecedented scale."

A major step toward a 'virtual mouse'

The new system could be used to study how genes affect areas across the body, or to study the effects of a new drug. "It can show how drugs are impacting tissues in ways that weren't predicted," Chevrier said. 

The next goal is to use the system to model not just one slice of a mouse but the whole mouse body. That's an important step to creating the kind of data which could one day help with creating a "virtual mouse" model that could be used in place of actual mice for research. 

"We think this data could be one of the enabling technologies to fulfill this vision of a virtual laboratory mouse model," Chevrier said. 

Other authors on the paper include Maggie Clevenger, Denis Cipurko, Ashwini Patil, Bohan Li, Michihiro Takahama, Linghan Mei, Madison Plaster, Gabriella Richey, and Feng Bao.