Project Summary/Abstract Complex diseases studies have revealed how specific cell types contribute to the evolution of different diseases. Many drugs are not effective to a large population of patients due to the cellular diversity. Modern single-cell RNA sequencing (scRNA-seq) technologies provide opportunities to detect and dissect the heterogeneity in cells, enable us to measure the gene expression level of thousands of individual cells in a single experiment. Comprehensive analysis of scRNA-seq data and correct reconstruction of cell-type-specific regulatory networks could help develop personalized and targeted treatment of some complex diseases for different patients. So, the scRNA-seq has more advantages than the traditional bulk RNA-seq. Most of genome and computational studies are based on the traditional bulk RNA sequencing data, which measure the average expression of the cell population, without examining the cell-type- specific expression profiles. There are several challenges in the scRNA-seq data analysis and cell-type-specific regulatory network reconstruction. The first challenge is, there are a large amount of missing values in the scRNA-seq data, which will attenuate the power and advantages of scRNA-seq, and make it difficult to correctly reconstruct a cell-type-specific network. We propose novel data-driven deep generative modeling methods to impute (estimate) the missing static and time-series scRNA-seq data without making certain distribution assumptions for the missing values. Some studies have revealed that the regulatory networks undergo systematic rewiring at different stages. It is of importance to know how many stages the cell has experienced, and when the stage transition starts to occur from the high- dimensional scRNA-seq data, which is the second challenge problem—change-points detection. We propose to develop an adversarial network-based method to identify the change-points without introducing model parameters. Another challenge is how to correctly reconstruct and intelligently validate the cell-type-specific regulatory networks from the scRNA-seq data, and identify key regulatory components that contribute to the network rewiring during stage transition. We propose to integrate the deep generative modeling methods and change-points detection algorithm with our weighted dynamic Bayesian network and Model Checking technique in a unified framework to reconstruct cell-type-specific regulatory networks. Our studies will improve our understanding of regulatory network dynamics, and provide a key to discovering the mechanisms underlying the pathogenesis of diseases.