Virus infections remain major threats to human health worldwide as demonstrated by COVID-19 pandemic caused by SARS-CoV-2 and its variants. All enveloped viruses must fuse with host membranes to initiate the infection process. Membrane fusion is a critical, but poorly understood biological process that is driven by protein conformational changes. Membrane fusion is a highly complex, multistage and multiscale process, which is difficult to investigate through scale-specific techniques (both experimentally and numerically). In this project, we propose to investigate the structural changes of fusion proteins and virus fusion through a combination of multiscale modeling, machine learning, and complementary experimentation. Because of our decades long experience in working with the herpes simplex virus (HSV), we will utilize the herpesvirus fusogen, gB, a class Ill fusion protein, as a model protein, to elucidate protein conformational changes during virus fusion. The specific research aims are: (1) To delineate the conformational changes of viral fusion proteins, through development of the machine learning facilitated enhanced sampling scheme for fusion proteins and delineation of the sequential conformation changes of gB protein for HSV fusion by a combination of machine learning and experiments; (2) To elucidate membrane fusion driven by viral fusion protein conformational changes, through development of a multiscale model for membrane fusion, assessment of the role of membrane fluidity in fusion and elucidation of the importance of the gB membrane proximal region on gB conformational changes and membrane fusion through combined simulations and experiments. This research will establish experimentally validated, powerful modeling platforms for exploration of the protein conformational changes and will bridge the multiple spatial and temporal scales involved in the fusion process. The machine learning method for enhanced sampling is highly innovative and crucial to capture the large-scale structural (conformational) changes and associated energy profiles of fusion proteins, and to identify the appropriate pathways during the fusion process. The integration of the gradient-based optimization method to the machine learning algorithm is novel to identify the most appropriate reaction coordinates.