Inferring single-cell trajectories via critical cell identification using graph centrality algorithm

Trajectory inference (TI) aims to infer cell differentiation trajectories in biological processes. Numerous computational methods have been developed to infer cell lineages from single-cell gene expression data. However, cluster-based methods involve a discretization that fails to capture the continuous nature of differentiation processes, while graph-based methods directly estimate the differentiation process from gene expression profiles without detecting subpopulations, making them susceptible to noise. To address these limitations, we propose scTICG, a single-cell trajectory inference method through critical cell identification and a greedy strategy. scTICG integrates the strengths of cluster-based and graph-based methods. Initially, a cluster-based backbone structure is constructed to serve as a coarse-grained trajectory. Then, considering the dynamics of cell state transitions and the influence of certain critical cells, we identify these critical cells using the graph centrality algorithm. Subsequently, these critical cells are leveraged to refine the trajectory using a greedy strategy. We evaluate scTICG on five public datasets and compare its performance with eight state-of-the-art trajectory inference methods. The experimental results demonstrate that scTICG can infer more accurate and robust trajectories compared to competitive methods. The R code for scTICG is freely available at https://github.com/DHUDBlab/scTICG.