Fig 6.

The training models for macrostates using different machine learning model: (a) Artificial neural networks, (b) decision tree, (c) random forest, and (d) one-vs-one random forest. All the training are based on 120,000 frames extracted from 12 μs MD trajectories. In each frame, 10,878 Cα pair-wise distances are used as features for model development. All models show high training and testing accuracy, validating the appropriation of machine learning models for VVD simulations.