On the crystallographic accuracy of structure prediction by implicit water models: Tests for cyclic peptides

Five small cyclic peptides and four implicit water models, were selected for this study. DEEPSAM, a structure prediction algorithm built upon TINKER, was used. Structures predicted using implicit water models were compared with experimental data, and with predictions calculated in the gas phase. The existence of very accurate X-ray crystallographic data allowed firm and conclusive comparisons between predictions and experiment. The introduction of implicit water models into the calculations improved the RMSD from experiment by about 13% compared with computations neglecting the presence of water. GBSA is shown to be consistently the best implicit water model.