Comparative Analysis of Algorithms for Extracting Frames from Unexplored Regions in Molecular Dynamics Simulations.

Molecular dynamics (MD) simulations are invaluable tools for studying molecular behavior, offering insights into the time-dependent evolution of molecular systems. However, these simulations often face significant limitations in exploring the full conformational space of the complex systems. The high dimensionality and the presence of energy barriers in the conformational landscape hinder the effective exploration of rare states, leaving large portions of the conformational space unexplored. To overcome these limitations and enhance the exploration of the conformational landscape, in this study, an attempt is made to develop an advanced framework that integrates time-lagged independent component analysis with several clustering algorithms, including K-means, K-medoids, kernel density estimation (KDE), and combinations such as isolation forest + K-means and isolation forest + K-means + KDE. Our approach systematically identifies unexplored regions of the conformational space from MD trajectories, enabling a more efficient and targeted adaptive sampling. This methodology enhances the exploration of sparsely populated conformational space, which is critical to understanding complex molecular processes. We evaluated the performance of these techniques in terms of computational efficiency, compatibility, and their ability to detect previously unexplored regions in the conformational landscape. The methodology was developed and tested using implicitly solvated pentapeptide trajectories available in the public domain. Our results show that isolation forest-based methods provide superior computational efficiency, while KDE-based techniques excel in resolving the unexplored conformational space with finer granularity. This work may contribute to the development of more effective adaptive sampling strategies, offering a comprehensive approach to exploring molecular conformational landscapes.