Molecular dynamics (MD) simulations are among the most powerful computational techniques and are widely used to explore the structural and dynamical properties of biomolecular systems. The massive data sets generated by these simulations, consisting of thousands of atoms evolving across millions of time steps, present major challenges for efficient and interpretable analysis. Traditional approaches such as dynamical network analysis (DNA) have been widely adopted to characterize correlated motions, but they are inherently limited to capturing linear dependencies. Consequently, essential nonlinear relationships underlying biomolecular function frequently remain undetected. To tackle this challenge, we developed a hybrid method that integrates DNA with deep learning, employing autoencoders to capture hidden patterns. Autoencoders are capable of reducing the dimensionality of MD trajectories while preserving essential features and uncovering complex, nonlinear correlations. By combining these with DNA, our method establishes a more comprehensive representation of biomolecular dynamics, capturing both linear and nonlinear modes of interaction. This integration allows the detection of nonlinear dependencies that conventional DNA cannot capture, while still maintaining interpretability. It provides a general framework that combines network analysis with deep learning to characterize complex dynamical features of molecular systems, including large-scale motions, inter-residue communication, and conformational transitions. To evaluate the effectiveness of the proposed approach, we applied it to MD simulations of the bacterial membrane insertase YidC, where the DDNA framework outperformed conventional DNA in capturing hidden and nonlinear dependencies.
Ghomi et al. (Sun,) studied this question.