Understanding how cells are organized within tissues is crucial to learning how organs function and how diseases like cancer or neurological disorders develop. Every tissue is made up of many different types of cells, and their physical arrangement influences how they communicate and work together. However, one of the biggest challenges in biology has been pinpointing exactly where individual cells are located within complex tissues.

A team of researchers from the University of Copenhagen has developed an innovative deep learning tool called SC2Spa that addresses this challenge by accurately mapping single-cell RNA sequencing data to their original spatial locations within tissues. Single-cell RNA sequencing allows scientists to measure gene activity in individual cells, but it often loses the spatial information about where those cells were in the tissue. Spatial transcriptomics techniques do preserve location information but sometimes at lower resolution.

Overview of SC2Spa

A SC2Spa is composed of a FCNN to map transcriptome to the spatial location. For training, SC2Spa uses the transcriptome from ST beads as the input and the spatial location of them on the ST assay as the output. The common genes within a specified range based on the Wasserstein distance are selected. The weights of the FCNN are constrained using L1 regularization to promoter sparsity. The trained FCNN is used to predict spatial location for the transcriptome from either ST beads or scRNA-seq data. The predicted location is further visualized on the ST assay. B The weights of the FCNN are used to identify SVGs

SC2Spa bridges this gap by using spatial transcriptomics data to teach the algorithm the complex relationships between gene expression and spatial position. This enables the model to predict where each single cell belongs within the tissue with remarkable accuracy. The researchers tested SC2Spa on different spatial transcriptomics platforms and found it consistently outperformed other existing methods. The tool is also robust across varying levels of spatial resolution, making it versatile for different experimental setups.

In addition to accurately locating cells, SC2Spa can identify spatially variable genes, genes whose expression changes depending on where a cell is situated in the tissue. Identifying these genes is important for understanding how the microenvironment influences cell behavior and function. For example, certain genes may be active only in cells at the edge of a tumor or near blood vessels.

By combining advanced RNA sequencing data with deep learning, SC2Spa offers researchers a powerful new way to explore tissues at single-cell resolution while keeping the crucial spatial context intact. This breakthrough will accelerate discoveries in many fields including cancer research, neuroscience, and developmental biology by providing a more detailed understanding of how cells organize, interact, and change in health and disease.

Availability – SC2Spa is available at https://github.com/linbuliao/SC2Spa.