Cell Trajectory Inference Based On Single Cell Stacked Auto Encoders

Single-cell RNA sequencing (scRNA-seq) has dominated research in recent years. It is a new technique for high-throughput sequencing of mRNA at the single-cell level. However, many challenges remain when analyzing single-cell transcriptome data, including the high dimensionality and data sparsity. Cells differentiate over time, and it is difficult to infer the state and type of cells at a point in time (dynamic change), and new questions arise as to when to analyze cells in greater depth and learn the trajectory of cellular continuity. The key to understanding cell behaviour lies in studying their dynamic change process through cell trajectory inference. To solve this problem, we propose scSAE (Single Cell Stacked Auto Encoder) a framework for cell trajectory inference of single-cell RNA data based on deep learning. The framework integrates a denoising self-encoder and a stacked self-encoder to learn cell-to-cell relationships in gene expression matrices. We compared scSAE with three other cell trajectory inference methods on three single-cell RNA sequencing datasets. scSAE outperformed existing cell trajectory inference tools and scSAE can be a powerful tool for analyzing single-cell RNA trajectory inference.