Cell clustering serves as a key task in transcriptomic data analysis, playing a crucial role in cell type annotation, marker gene identification, and the discovery of rare cell populations. With the advancement of single-cell RNA sequencing (scRNA-seq) technologies, accurate unsupervised clustering can significantly reduce the manual effort involved in cell type annotation and is essential for uncovering novel biological insights.

Recently, Hongwei Li’s team from China University of Geosciences published a research article titled “scSCC: A swapped contrastive learning-based clustering method for single-cell gene expression data” in the journal Quantitative Biology, proposing a novel clustering method combining instance contrastive learning and swapped contrastive learning for scRNA-seq data.

As illustrated in Figure 1, the research team developed the scSCC (Swapped Contrastive Clustering for scRNA-seq Data) framework, which leverages an instance contrastive module to learn cell representations and incorporates a swapped prediction strategy to inject clustering signals into the embedding space, leading to enhanced performance in the cell clustering task. Comprehensive experiments on real datasets demonstrate that scSCC can achieve more accurate and reasonable clustering performance in most cases. By projecting cell representations to the 2 dimensional t-distributed stochastic neighbor embedding (t-SNE) space, results of scSCC form clear boundaries between different clusters and reflect more separable structure. Compared with ablated versions of scSCC, the entire scSCC model achieves best comprehensive performance, exhibiting the effectiveness of the swapped prediction module.