“…ML FFs hold promise, but regrettably, they are not yet fully developed for general applications. They are facing essential challenges related to instabilities, representation issues, and limitations in transferability, particularly when applied to larger systems. ,, However, their rapid development cycle is encouraging, setting them apart from traditional FFs, with numerous new ML FFs being introduced annually. ,,,,, For instance, Fu et al have provided valuable insights into the limitations of current training methods for ML FFs, which are likely to inspire further advancements in the field . The atomistic ML FFs hold great promise in achieving quantum mechanics-level accuracy in MD simulations, with the potential to introduce greater flexibility in handling configurational and conformational variations in small molecules, peptides, and proteins.…”