Single-cell RNA sequencing, or scRNA-seq, has given scientists an unprecedented ability to look at gene expression in individual cells. Instead of seeing an average across millions of cells, researchers can now explore the unique patterns that make each cell type distinct. This has opened exciting possibilities in areas like cancer research, immunology, and developmental biology. But there is a catch. scRNA-seq data is messy, noisy, and incredibly complex. Making sense of it requires powerful computational tools that can separate different cell types and identify their unique markers.

A team led by researchers from the Chinese Academy of Sciences has introduced a new approach to address this challenge. Their framework, called scSiameseClu, is designed to improve how researchers cluster cells based on RNA sequencing data. Clustering is a vital step because it helps scientists categorize cells into groups, revealing their identities and relationships.

Overview architecture of scSiameseClu

It contains three components: (i) a data-augmented module, (ii) a Siamese fusion module, and (iii) an optimal transport clustering strategy for self-supervision learning.

So what makes scSiameseClu different? The framework works in three main steps. First, it uses a Dual Augmentation Module, which introduces biologically informed variations into the data. By slightly “perturbing” both the gene expression matrix and the relationships between cells, the system learns to be more robust against noise and errors.

Next, it applies a Siamese Fusion Module. This clever step refines correlations between datasets and fuses information in a way that prevents a common issue in machine learning called “over-smoothing.” Over-smoothing happens when models average out too much detail, losing the subtle but important differences between cell types.

Finally, the Optimal Transport Clustering step uses a mathematical approach called the Sinkhorn distance to balance how cells are assigned to clusters. This ensures that the clusters are not only accurate but also efficiently aligned with expected proportions.

When tested on seven real-world datasets, scSiameseClu performed better than current state-of-the-art methods in clustering cells, annotating cell types, and classifying them. This means it can help researchers interpret RNA sequencing data more accurately and with greater confidence, potentially speeding up discoveries in biomedical research.