Using Spectral Clustering to Sample Molecular States and Pathways

Molecular dynamics (MD) simulations offer a way to explore the conformational state space of large, biologically relevant molecules. Our sampling method, called “concurrent adaptive sampling” (CAS), utilizes MD simulations by letting a number of “walkers” adaptively explore the state space in consecutive steps. Walkers start in one conformational state, execute a short MD simulation and thus end up in a different state. One of the properties of CAS algorithm is that the number of walkers explodes quickly as the full state space of the biomolecule is explored. Hence, we use a technique called “spectral clustering” in order to cluster similar walkers whenever there are too many walkers in the state space. After clustering, multiple similar walkers are replaced by a lower number of walkers within each cluster. This results in a large speedup of the sampling algorithm.