Understanding how cells are organized within tissues is a major focus in biology, but the tools scientists use often come with limitations. Spatial transcriptomics is a powerful method that shows where genes are active in tissue samples, yet it usually captures only a fraction of gene activity, leaving many details unknown. To address this challenge, a team led by researchers from the University of Florida have developed SpaIM, a new computational model that combines spatial transcriptomics with single-cell RNA sequencing data. The researchers created SpaIM to better predict missing gene activity in spatial maps, allowing scientists to see a fuller picture of cellular function.

SpaIM works by using a technique called style transfer learning, which separates shared biological signals from dataset-specific patterns. By doing so, the model integrates the richness of single-cell RNA sequencing with the positional information provided by spatial transcriptomics. In tests across 53 datasets representing both sequencing- and imaging-based spatial technologies, SpaIM outperformed 12 other leading methods, improving gene coverage and accuracy.

Overview of the SpaIM model

SpaIM comprises an ST autoencoder and an ST generator. Both the ST autoencoder and the ST generator are built on the multilayer recursive style transfer (ReST) layers.

Importantly, SpaIM is not only about filling in missing data. By providing more complete gene information, it improves several downstream applications such as identifying cell-to-cell communication through ligand-receptor interactions, defining spatial domains in tissues, and detecting changes in gene activity between conditions. Because SpaIM is open-source software, researchers worldwide can use it to advance their studies in tissue biology, disease research, and precision medicine.

This work highlights how combining RNA sequencing with spatial data can lead to a deeper understanding of how cells interact within tissues, ultimately helping us explore the complex architecture of life at a much finer scale.

Availability – All source codes and data in our study have been deposited at https://github.com/QSong-github/SpaIM.