A Network Propagation Based Approach for Measuring Cell-Cell Similarity

The rise of single-cell RNA sequencing (scRNA-seq) technology allows us to capture transcriptomic data at the individual cell level. Utilizing unsupervised clustering based on transcriptome similarity aids in accurately identifying essential cell types. However, challenges such as high dimensionality, noise, and data sparsity within scRNA-seq pose difficulties for conventional similarity assessment. Addressing these challenges is crucial for understanding heterogeneity among cell populations. Consequently, we propose a novel approach, CellSim, for assessing cell-cell similarity based on network propagation. We evaluated the method’s performance by clustering cells across 11 scRNA-seq datasets. Simultaneously, we conducted simulation experiments to assess the method’s ability to identify relationships between cells in datasets with high dropout rates. The results indicated that CellSim is superior to others in terms of evaluating clustering results. Eventually, we apply these similarity methods to the same trajectory inference framework on embryo datasets. The results reveal that our method outperforms others in inferring the cell trajectory and exhibits strong consistency with the ground truth.