This paper in Journal of Cheminformatics reviews how deep learning (DL) is being integrated into molecular modelling, particularly in virtual screening (VS) and molecular dynamics (MD) simulations, to enhance drug discovery workflows. DL techniques are shown to improve accuracy, speed, and analysis of simulations by addressing limitations in traditional computational chemistry, such as hardware constraints and algorithmic inefficiencies. The review is structured around four key areas: DL-enhanced VS workflows, DL-guided MD simulations, neural network-based force field approximations, and DL-driven MD trajectory analysis. A broad range of DL models are presented —including convolutional neural networks, graph neural networks, and generative models like GANs and VAEs—alongside real-world applications such as DEEPScreen, DiffDock, and AlphaFold. The paper concludes that DL holds the potential to transform molecular modelling, but also notes challenges like data quality, model interpretability, and the need for broader accessibility and standardization.