Reconstruction of Cellular Trajectories with Biologically Interpretable States

Reconstructing cellular trajectories offers biologists a more comprehensive understanding of biological systems and disease progression, enabling the discovery of crucial mechanisms and the design of effective therapies. Single-cell RNA sequencing (scRNA-seq) is a powerful technique used to investigate cellular evolution. However, scRNA-seq presents two significant challenges.

Firstly, the process involves cell destruction during gene expression measurement, making it impossible to monitor the gene-expression profile of the same cell throughout its entire trajectory. Secondly, since scRNA-seq captures cells at discrete time points, it only provides static snapshots instead of continuous trajectories, restricting its ability to fully capture the dynamics of cellular transitions.

To tackle these issues, researchers have developed various computational methods. However, current models fall short in capturing the hidden state of each cell, which is crucial for reconstructing gene expression at unobserved time points. The expression profiles of cells are influenced by their state and can change over time. In response, we introduce the Latent Cell-State Model.

This model approximates cell state transitions as a Markov chain and derives the gene expression of each cell at unobserved timepoints based on the cell state, time, and a latent variable that represents the high-dimensional gene expression profile of a cell. Our model excels at inferring biologically meaningful cell states and generating high-quality gene expression profiles at unobserved timepoints, demonstrating its potential for accurate trajectory reconstruction.