Single-cell approaches provide insights into individual cells, enabling measurement of gene expression and chromatin accessibility (Peaks). However, understanding how they influence each other remains a central challenge, since regulatory interactions span long genomic distances, as well as the sparsity and high dimensionality of single-cell data. This thesis investigates the use of Variational Autoencoders (VAEs) to model the relationships between gene expression and chromatin accessibility. The single-cell matrices of gene expression and peaks were obtained from the PBMC 3k and PBMC 10k datasets (10x Genomics), which relate to human immune cells. Three VAE-based architectures were implemented, including the main Gene–Peak–Gene (GPG) model, a reverse Peak–Gene–Peak (PGP) model, and a dual VAE with a shared loss term.