SARS-CoV-2 infection proceeds through the binding of viral surface spike protein to the human ACE2 protein. The global spread of the infection has led to the emergence of fitter and more transmissible variants with increased adaptation both in human and nonhuman hosts. Molecular simulations of the binding event between the spike and ACE2 proteins offer a route to assess potential increase or decrease in infectivity by measuring the change in binding strength. We trained a neural network model that accurately maps simulated binding energies to experimental changes in binding strength upon amino acid changes in the spike protein. This computational workflow can be used to a priori assess currently circulating and prospectively future viral variants for their affinity for hACE2.
The association of the receptor binding domain (RBD) of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike protein with human angiotensin-converting enzyme 2 (hACE2) represents the first required step for cellular entry. SARS-CoV-2 has continued to evolve with the emergence of several novel variants, and amino acid changes in the RBD have been implicated with increased fitness and potential for immune evasion. Reliably predicting the effect of amino acid changes on the ability of the RBD to interact more strongly with the hACE2 can help assess the implications for public health and the potential for spillover and adaptation into other animals. Here, we introduce a two-step framework that first relies on 48 independent 4-ns molecular dynamics (MD) trajectories of RBD−hACE2 variants to collect binding energy terms decomposed into Coulombic, covalent, van der Waals, lipophilic, generalized Born solvation, hydrogen bonding, π−π packing, and self-contact correction terms. The second step implements a neural network to classify and quantitatively predict binding affinity changes using the decomposed energy terms as descriptors. The computational base achieves a validation accuracy of 82.8% for classifying single–amino acid substitution variants of the RBD as worsening or improving binding affinity for hACE2 and a correlation coefficient of 0.73 between predicted and experimentally calculated changes in binding affinities. Both metrics are calculated using a fivefold cross-validation test. Our method thus sets up a framework for screening binding affinity changes caused by unknown single– and multiple–amino acid changes offering a valuable tool to predict host adaptation of SARS-CoV-2 variants toward tighter hACE2 binding.