Constructing metastable states networks in high-dimensional conformational space can greatly improve the understanding of complex molecular systems, such as DNA an proteins and furthermore enhance efficient of simulations.

The high-dimensional free energy landscape and dynamical properties of systems can be honestly reproduced with the network representation without requiring a priori assumptions about reaction coordinates.  We develop a method for naturally reconstruct the hierarchical transition networks among metastable states.  Multiple (short) simulation trajectories are generated in parallel, and each trajectory is mapped into a high-dimensional vector with the averages of lots of conformational functions along the trajectory as its components.  The linear space spanned by the trajectory-mapped vectors has the same structure as that spanned by the conformational probability density functions of these trajectories, thus simple linear algebraic analyses on the mapped vectors can identify metastable states, transition kinetics as well as transition pathways of the simulation trajectories.  The method is useful in data analysis of high-dimensional time sequences. and comparison of two experimental samples.  We illustrate the application of the method to understand folding-unfolding dynamics and mechanisms of polypeptides.