LISA: Accurate reconstruction of cell trajectory and pseudo-time for massive single cell RNA-seq data.

Cell trajectory reconstruction based on single cell RNA sequencing is important for obtaining the landscape of different cell types and discovering cell fate transitions. Despite intense effort, analyzing massive single cell RNA-seq datasets is still challenging. We propose a new method named Landmark Isomap for Single-cell Analysis (LISA). LISA is an unsupervised approach to build cell trajectory and compute pseudo-time in the isometric embedding based on geodesic distances. The advantages of LISA include: (1) It utilizes k-nearest-neighbor graph and hierarchical clustering to identify cell clusters, peaks and valleys in low-dimension representation of the data; (2) Based on Landmark Isomap, it constructs the main geometric structure of cell lineages; (3) It projects cells to the edges of the main cell trajectory to generate the global pseudo-time. Assessments on simulated and real datasets demonstrate the advantages of LISA on cell trajectory and pseudo-time reconstruction compared to Monocle2 and TSCAN. LISA is accurate, fast, and requires less memory usage, allowing its applications to massive single cell datasets generated from current experimental platforms.