Sequence-dependent conformational sampling using a database of phi(i)+1 and psi(i) angles for predicting polypeptide backbone conformations

A method to sample the conformational space of polypeptides in a sequence-dependent manner was developed and applied to predict backbone conformations of polypeptides. The method uses a database of phi(i)+1 and psi(i) angles sorted into pools corresponding to 400 dimers of naturally occurring amino acids. The pools of phi(i)+1 and psi(i) angles for dimers were further dynamically divided into sub-pools based on the homology of two amino acids on either side of the dimer. Typically, 10000 backbone conformations for every overlapping hexamer of a polypeptide are generated by randomly assigning a set of phi(i)+1 and psi(i) angles for every dimer from the sorted sub-pool for that dimer. The conformational preference for each hexamer is evaluated by superimposing each simulated conformation on a set of pre-selected hexameric templates which are derived from high-resolution crystal structures representing helical, extended and turn conformations. A model is classified as compatible with a template if it has a backbone root mean square deviation of < or = 1 A to that template. The template that has the highest population of compatible models for a given hexamer is interpreted as the most probable conformation of that hexamer. The method was applied to predict backbone conformations that arise primarily from local interactions and the predictions were compared with experimental results. Results from simulations on structurally characterized polypeptides in helical, extended, turn and mixed alphabeta conformations are presented to demonstrate the predictive powers of the method. It is also demonstrated how models of overlapping hexamers can be assembled to obtain near-native models for the entire sequence.