Dynamic insights into the structural evolution of ACE2-RBD interactions through molecular dynamics simulation, Markov state modeling, and large language model mutation prediction.

The viral Spike protein of SARS-CoV-2 is responsible for virus-host recognition, in which the receptor-binding domain (RBD) binds to the host receptor angiotensin-converting enzyme 2 (ACE2). Structural biology studies have revealed the structures and binding mechanisms of several variants of concern, but they inherently lack dynamic information. Previous molecular dynamics (MD) simulations have generated vast trajectories, but the metastable states are insufficiently detected and analyzed. In this study, we dissect extensive MD simulations of the ACE2-RBD complex of the wild type and three pivotal Omicron variants: BA.2, BA.2.86, and JN.1. Employing the Time-lagged independent component analysis-Markov State Models method, we identify four principal metastable states for each system and demonstrate that lineage-defining mutations systematically alter the thermodynamic properties, such as equilibrium populations and interaction energies, as well as their transition kinetics. Our study establishes a robust paradigm for extracting detailed thermodynamic and kinetic insights from MD simulations that surpass conventional analysis. Furthermore, by scanning point mutations on the RBD, evaluating these mutations with a fine-tuned large language model, and investigating the resulting structural interactions at the binding interface, our work provides a dynamic, atomic-level view of the structural evolution of ACE2-RBD interactions, offering a mechanistic basis to aid in future viral surveillance and therapeutic design.