Adaptive Normal Mode Sampling (aMDeNM) Enhances Exploration of Protein Conformational Space and Reveals the...

Proteins exhibit a diverse range of structures and dynamics that are critical to their biological function. These dynamic processes span a broad spectrum of time scales and are influenced by environmental factors, including temperature, solvent composition, and the presence of binding partners or membranes. Efficient exploration of protein conformational space is essential for understanding their functional mechanisms, but this remains challenging because of the high dimensionality of the energy landscape. Our group has previously developed the molecular dynamics with excited normal modes (MDeNM) method, which is based on the kinetic excitation of normal modes (NMs) during molecular dynamics simulations. Here, we developed an adaptive extension of the method (aMDeNM), where the motions described by preselected directions of low-frequency NMs are dynamically adjusted throughout the simulation. By coupling low-frequency NM excitation with adaptive directional adjustments, aMDeNM facilitates extensive exploration of the energy landscape, overcoming the constraints of fixed, rectilinear displacements and alleviating structural stresses and environmental resistance. The method was tested on three structurally diverse test systems: T4 lysozyme, human calmodulin, andStaphylococcus aureus monofunctional transglycosylase. Our results demonstrate improved conformational sampling compared with standard MD and other enhanced sampling methods. Additionally, spectral analysis of structural oscillations along the pathways using fast Fourier transform revealed the role of low-frequency vibrations in critical conformational changes and highlighted the influence of the surrounding environment on protein dynamics. This work provides a robust framework for studying large-scale protein motions and their functional implications within complex biological environments. Importantly, aMDeNM requires only an initial structure without the need to specify predefined target states, distinguishing it from many biased sampling techniques that rely on predefined target conformations. The aMDeNM code and usage instructions are available at https://github.com/pedro-tulio/amdenm.