From Structure to Function: Predicting Protein Conformational Changes

Proteins are high-dimensional dynamical systems with tens of thousands of internal degrees of freedom, yet functional conformational changes typically proceed along only a few highly collective coordinates. Predicting these motions remains challenging because the time scales of functional transitions—often milliseconds to hours—are far beyond those accessible to conventional molecular dynamics simulations. Here we present a rigorous physics-based framework based on potential energy flow and the generalized work functional, which enables direct computation of the true reaction coordinates governing protein conformational changes from short energy-relaxation simulations starting from the native structure. These coordinates enable predictive simulations that generate unbiased dynamical trajectories of conformational transitions and identify functionally important conformational states using the native structure as the sole input. We demonstrate the general applicability of this approach across diverse systems, including the PDZ domain, human major histocompatibility complex I, tubulin, and HIV reverse transcriptase. These results suggest that the information governing long-timescale functional transitions is encoded in the native-state energy landscape, enabling prediction of hidden conformational states directly from protein structure.