Understanding the epigenetic mechanisms that regulate cell fate and function requires deciphering how chromatin accessibility patterns define gene regulatory networks (GRNs). High-resolution single-cell ATAC-seq technologies enable the identification of open chromatin regions at single-cell resolution, providing a unique opportunity to reconstruct cis-regulatory interactions. However, inferring GRNs from such data remains challenging due to extreme sparsity, high dimensionality, and complex chromatin organization. Current computational tools—such as CellOracle, SCENIC+, and STREAM—mainly rely on enhancer–promoter interactions to infer GRN, while the contribution of insulators and other architectural elements to chromatin topology and gene regulation remains underexplored. This project aims to develop a computational framework for inferring GRNs from single-cell chromatin accessibility data and single-cell transcriptomic data by jointly modeling enhancers, promoters, and insulators. Specifically, the method will integrate sequence-derived features, 3D chromatin conformation priors, and co-accessibility patterns using machine learning approaches, such as graph neural networks (GNNs) and variational autoencoders (VAEs), to capture both local and global dependencies in the chromatin landscape. The framework will be benchmarked on publicly available datasets. The project will require significant computational resources for large-scale data preprocessing, model training, and network inference. GPU acceleration will be essential for deep learning model training, while CPU nodes and high-memory nodes will be used for data integration, feature extraction, and downstream statistical analyses. The anticipated outcome is a generalizable and interpretable computational method that advances our understanding of how the chromatin regulatory architecture shapes cell-type–specific transcriptional programs.