In our latest research, we present advancements in modeling protein dynamics using coarse-grained potentials based on artificial neural networks. We leverage a multi-millisecond dataset of molecular dynamics (MD) simulations to train a coarse-grained potential that can model protein dynamics on computationally accessible timescales, recover the native structure of twelve fast-folding proteins and other non-native metastable states, as well as folding pathways and the formation of various secondary and tertiary protein structures.
Furthermore, we also construct a single coarse-grained potential trained with all twelve proteins and accurately reproduce experimental structural characteristics in some mutated versions of the proteins. This research suggests that machine learning-based coarse-grained potentials hold promise as a viable approach for simulating and comprehending protein dynamics, addressing a longstanding challenge in the field.
You can read more about this here.