Single-cell RNA sequencing (scRNA-seq) has revolutionized how scientists understand the complexity of tissues by examining gene activity in individual cells. However, the data generated from scRNA-seq experiments is often noisy and high-dimensional, making it hard to group similar cells accurately, a process known as clustering.

Researchers at Anhui University have developed a new computational method called scMCGF to tackle this challenge. This innovative algorithm uses a technique known as multi-view graph fusion to combine different types of information about cells and build a unified, high-quality representation of how similar cells are to one another.

Specifically, scMCGF brings together data from multiple perspectives, including gene expression patterns, pathway activity scores, and dimensionality-reduction outputs like PCA and diffusion maps. It then creates and refines similarity graphs for each data view and fuses them into a single, unified graph. This fused graph is used to cluster the cells more accurately.

Overview of the scMCGF workflow

(a) Multi-view data construction. We pre-process the raw scRNA-seq data and calculate its pathway score matrix to form two initial data views (View1 and View2). Subsequently, PCA and diffusion maps are utilized to extract linear and nonlinear features, respectively. These extracted features constitute two additional views (View3 and View4). (b) Multi-view fusion learning and clustering. The similarity graphs of each view (S1 ~ S4) and the unified graph (U) are learned jointly through adaptive weighted fusion. The final clustering results are obtained by imposing a rank constraint on the Laplacian matrix of the unified graph U.

The researchers tested scMCGF on 13 real-world scRNA-seq datasets. Their results show that scMCGF outperforms eight leading clustering tools in both accuracy and stability. Importantly, the method also grouped cells in ways that matched known biological patterns, making it a strong foundation for future research in cell biology, cancer, and developmental biology.

Availability – The source code is available at https://github.com/WJ319/scMCGF.