Although the detailed structures of proteins in their native (fully folded and functional) states are crucial for designing drugs and understanding how proteins work, researchers are increasingly recognizing the importance of protein intermediate states—also known as “folding intermediates,” “dark states,” “excited states,” or “dynamic states.” These intermediate states are formed as proteins fold or undergo conformational transitions. Previous reports have assigned significant functional roles to these states, including serving as potential drug targets. However, detecting these conformational states is challenging and often requires advanced techniques such as relaxation dispersion NMR spectroscopy, time-resolved cryo-electron microscopy, fluorescence-based lifetime assays, and advanced trajectory analysis in simulations.
A major hurdle in computationally identifying these intermediate states is the requirement of an enormous amount of computer time. Recently, Gerstein and co-workers developed a new molecular dynamics (MD) algorithm designed to address this issue. Their MD algorithm, called the “discard-and-restart,” significantly speeds up the process by systematically discarding short simulations that do not yield useful information and restarting them with new atomic velocities drawn from a Maxwell–Boltzmann distribution (a statistical method). The decision to continue, discard, or restart a simulation is guided by a collective variable loss function—a mathematical tool that measures how far a simulation’s outcome is from a desired or target property, helping select promising simulation paths. Through these iterative cycles, the algorithm achieves up to a 2,000-fold increase in speed compared to traditional MD approaches, while still capturing key features of protein folding at a fraction of the computational cost.
This innovative approach allows researchers to rapidly sample and observe transient intermediate states of proteins at the atomic level, including pathways of folding and partial unfolding, without compromising structural accuracy. By making it feasible to explore a broader range of protein conformations, the algorithm greatly expands the set of structures that can be considered in structure-based drug design. In drug discovery, rapid detection of these intermediate states is particularly valuable because it enables the identification of new binding sites or unique conformations that may serve as novel drug targets but would otherwise remain hidden in traditional studies focused only on the stable, native state.
This newly developed MD algorithm has been successfully tested on several fast-folding small protein domains, such as the TrpCage miniprotein, Fip35 WW domain, villin subdomain, and β hairpin fragments. By enabling faster and more comprehensive detection of protein intermediate states, this algorithm opens new avenues for researchers to discover previously inaccessible conformations and binding sites, thereby accelerating the design and development of innovative therapeutics.