Evaluating elastic network models in prediction of conformational changes of proteins

Conformational changes during protein -protein or ligand -protein docking play an important role in the biological processes. These changes involve low frequency collective motions, and normal mode analysis is generally used for finding the frequencies and mode shapes of the proteins. Many studies have been focused on the prediction of these transitions using different protein models. Among them, elastic network models are popular, as they are simple and do not require energy minimization. However, so far no systematic study has been done about considering the effect of different parameters in prediction of these conformational changes. In this study 20 proteins with pre-determined conformational changes were selected and the success and validation of each elastic network model in predicting the bound state were tested. According to the results, the first three modes play the major role in predicting the conformational changes. Moreover, choosing the proper cutoff radius is more effective than the potential function. Results also show that non-exponential models with 10 angestrom cutoff are more accurate in predicting conformational changes, in spite of their simplicity and being less time consuming.