Spatial transcriptomics (ST) is an innovative technique that allows researchers to study gene activity within the context of tissue architecture. This technology helps scientists understand how cells interact and form complex structures within tissues. However, a challenge has emerged with some of the current ST technologies—many operate at low resolution, which means that a single data point, or “spot,” may contain multiple, distinct cell types. This can make it difficult to pinpoint specific cellular details and understand how these cells contribute to the overall tissue structure.

In addition, there are limitations in the current computational methods, known as spatial deconvolution, that are used to interpret this complex data. These methods often fail to fully utilize other types of biological information, such as gene expression combined with chromatin accessibility (the ability of DNA to be accessed for transcription).

Researchers at Wuhan University have developed CLPLS (Contrastive Learning and Partial Least Squares regression), a cutting-edge computational tool designed to improve the analysis of ST data. By using CLPLS, researchers can more accurately break down the “spots” in ST data into their individual cell types, even when those spots contain a mix of different cells. But what sets CLPLS apart from other methods is its ability to integrate additional data, like chromatin accessibility, which provides a deeper insight into how genes are regulated within cells. This multi-modality approach allows for a more complete understanding of the cellular behavior in tissues.

The overview of CLPLS

The inputs of CLPLS are spatial transcriptomics data, scRNA-seq and paired scATAC-seq data (if available). The low-dimensional embeddings are obtained by the representation learning module and are then inputted into the PLSR module. The output of CLPLS is a spot-by-cell matrix, which can be used for estimating cell type compositions.

The team behind CLPLS applied this method to both simulated and real datasets, and their results showed that CLPLS outperformed other methods in deconvoluting data at the single-cell level. By improving how we interpret spatial transcriptomics, CLPLS paves the way for new discoveries in tissue biology, offering a more precise view of how gene expression and cellular structures work together in health and disease.

This method holds great potential for advancing research into complex tissues, like tumors or organs, where understanding the spatial arrangement of cells is crucial for uncovering disease mechanisms and developing targeted therapies.

Availability – The source code of CSPLS is available at https://github.com/LindsayMo/CLPLS_package.