Identification of crucial interactions dictating membrane fusion through machine learning

Enveloped viruses must enter host cells to initiate infections through a fusion process, during which the fusion proteins undergo significant and complex structural changes. Understanding of the fusion protein conformational stability, rapid and accurate identification of the stabilizing interactions are critically important for inhibiting the infections. In this work, we leverage molecular dynamics simulations and machine learning models to identify the crucial interactions dictating the structural stability of fusion protein. We demonstrate our method using the glycoprotein B (gB), a class III herpesvirus fusion protein, as a model. Using our method, we extensively evaluated the interactions between the fusion loops and the membrane proximal region in gB, based on which a new interaction was identified from our machine learning model as critical in stabilizing gB. Molecular simulations revealed that mutation of one of the interacting residues disrupted the fusion loop secondary structure and reduced gB stability. In addition, experiments were conducted to evaluate the impact of the identified interaction. Strikingly, the mutation completely abrogated gB membrane fusion activity, which is consistent with the model predictions. The results deepen our fundamental understanding of the molecular mechanisms of viral fusion, which may lead to novel antiviral interventions. The modeling can be generalized to rapidly identify the critical inter-molecular interactions in other important biological processes.