Publication
We introduce a novel approach for discoveringeffective degrees of freedom (DOF) in molecu-lar dynamics simulations by mapping the DOFto approximate symmetries of the energy land-scape. Unlike most existing methods, we donot require trajectory data but instead rely onknowledge of the forcefield (energy function)around the initial state. We present a scalablesymmetry loss function compatible with exist-ing force-field frameworks and a Hessian-basedmethod efficient for smaller systems. Our ap-proach enables systematic exploration of confor-mational space by connecting structural dynam-ics to energy landscape symmetries. We applyour method to two systems, Alanine dipeptideand Chignolin, recovering their known importantconformations. Our approach can prove usefulfor efficient exploration in molecular simulationswith potential applications in protein folding anddrug discovery.
NetSI authors
