While there are theoretical virtues and BMY-14802 limitations of the adaptive T-ReX method, we investigate in this work the practical performance of the method applied to the structure refinement of a dataset of protein conformational decoys. Given the success of the adaptive T-ReX simulations for modeling the sharp energy differences between the folded and unfolded states of SH3, it is of general interest to determine whether this approach is beneficial in modeling less-cooperative transitions that are thought to govern the structure refinement of decoys taken from either comparative modeling or knowledge-based structure predictions. We examine a set of protein targets that offers a challenging benchmark of simulation methods for structure refinement. The targets are 49�C92 residues in size and amenable to an analysis by unrestrained all-atom simulation methods. Our simulation protocol is identical to that applied in earlier studies of refinement. Computational sampling is conducted by combining the selfguided Langevin dynamics simulation method with replica-exchange simulations, and we provide a comparison between the application of dynamic client walkers in temperature space and that of using a static distribution of temperatures. We also investigate four different energy functions to rank order conformations from the simulations. Of the multiple metrics to assess the accuracy of the simulations, the fraction of native contacts is perhaps one of the more stringent structural measures of native-like conformations. Quality assessment of the modeled structures can be further evaluated by the customary use of Ca root-mean-square deviation ; however, conformers rank ordered by this measure may contain poor side-chain packing. From the overall results, the adaptive TReX simulations for the aggregate dataset gives an average fN value of 0.62, computed as a AL-8810 statistical average over the four energy methods using the top-scoring conformer detected by each energy function. As a comparison, the T-ReX simulations using a static temperature condition yielded a nearly equivalent average of fN 0.60. The starting decoys for the targets prior to refinement have an average fN of 0.51 for the top-ranked conformers using the RWplus energy function to rank-order structures. The overall range for the starting decoy set is a top value of fN 0.64 to a low value of 0.29.