TIPD: A Probability Distribution-Based Method for Trajectory Inference from Single-Cell RNA-Seq Data

Single-cell RNA-seq technology provides an unprecedented opportunity to allow researchers to study the biological heterogeneity during cell differentiation and development with higher resolution. Although many computational methods have been proposed to infer cell lineages from single-cell RNA-seq data, constructing accurate cell trajectories remains a challenge. We develop a novel trajectory inference method-based probability distribution (TIPD) to describe the heterogeneity of cell population. TIPD combines signalling entropy and clustering results of the gene expression profile to describe the probability distributions of heterogeneous states in a cell population. It does not require external knowledge to determine the direction of the differentiation trajectories, so its application is not limited by the annotations of the data set. We also propose a new distance metric to measure the distance of the probability distributions of the identified heterogeneous states. On this distance matrix, a minimum spanning tree (MST) is built to reorganize the order of cell clusters. The constructed MST is calculated based on systems-level information, so it is consistent with the real biological process. We validated our method on four previously published single-cell RNA-seq data sets including the linear structure and branch structure. The results showed that TIPD successfully reconstructed the differentiation trajectories that are highly consistent with the known differentiation trajectories and outperformed the other four state-of-the-art methods under different assessment criteria.