Bridging Molecular Simulations and Experimental Observables: Predicting FRET Signals and Conformational Transitions Using MD and Deep Learning in Cas9

This study applies machine learning to predict single-molecule Förster Resonance Energy Transfer (smFRET) efficiency from structural features derived from molecular dynamics simulations. Using Cas9 as a model system, inter-residue distances, center-of-mass displacements, radius of gyration, interdomain angles, and linker flexibility were extracted and used to train Hidden Markov Models and Long Short-Term Memory networks. These models were applied to characterize transitions between conformational states associated with Cas9 activation. Predicted FRET efficiencies and state transitions were compared to experimental smFRET data, with qualitative similarities in transition patterns and histograms. Due to differences in simulation and experimental timescales, direct comparison of kinetics was not performed. This approach explores the relationship between simulated structural dynamics and FRET-relevant observables, providing a framework for further studies that integrate computational models with experimental conformational analysis. The method may contribute to understanding how structural features influence conformational transitions in dynamic proteins.