Time-Course Single-Cell Multimodal Analysis and Trajectory Inference Using Deep Generative Models

• Activity Number: 496 - Machine Learning Methods for Single-Cell Analysis
• Type: Invited
• Date/Time: Thursday, August 11, 2022 : 8:30 AM to 10:20 AM
• Sponsor: WNAR
• Author(s): Qiao Liu* and Xi Chen and Jingxue Xin and Wanwen Zeng and Wing Hung Wong
• Companies: Stanford University and Stanford University and Stanford University and Stanford University and Stanford University
• Keywords: single-cell; deep learning; trajectory inference; multi-omics; generative model; neural network

Abstract

Single-cell multimodal data with time-course information could reveal the dynamic changes in both trans- and cis-regulatory activities during continuous biological processes. It is crucial to infer the cell state transition underlying the dynamic regulatory mechanism along the cell developmental trajectories. The majority of computational methods aim at inferring the pseudo-temporal trajectories within a single biological sample or analyzing gene activity dynamics based on singleomic data (e.g., scRNA-seq). Here, we proposed scDEC-TI, a computational tool for time-course single-cell multimodal analysis and trajectory inference, based on deep generative neural networks. scDEC-TI is a two-stage model and is capable of integrative single-cell analysis and inferring the major trajectories during the cell development. With our newly sequenced 15,514 cells with both scRNA-seq and scATAC-seq profiles, we demonstrate the advantages of scDEC-TI in decoding the dynamics regulatory programs during the definitive endoderm differentiation.