Analyzing dynamical simulations of intrinsically disordered proteins using spectral clustering

Continuing improvements in algorithms and computer speeds promise that an increasing number of biomolecular phenomena can be simulated by molecular dynamics to produce accurate "trajectories" of their molecular motions on the nanosecond to microsecond time scale. An important target for such simulations will be non-equilibrium biochemical processes, such as protein folding, but existing tools for analyzing molecular dynamics trajectories are not well suited to non-equilibrium processes and progress will require improvements in tools for classifying the range and types of dynamics exhibited by these systems. An extreme example of a non-equilibrium biochemical process is the function of "intrinsically disordered" proteins - proteins that function without ever folding into a unique structure. In this paper, we demonstrate the use of spectral clustering methods to analyze the data produced from simulations of several forms from one class of intrinsically disordered proteins, the phenylalanine-glycine nucleoporins (FG-Nups). We explain why such methods are well-suited for the data produced by our simulations and show that clustering methods provide a direct, quantitative measure of how effectively single simulations independently sample regions of structural phase space. Moreover, our clustering results show distinct dynamical behavior in different forms of the FGNups, which may provide insights into their biological function.