On the application of phase relationships to complex structures. XXX. Ab initio solution of a small protein by SAYTAN

An account is given of experiments with the program SAYTAN to determine directly the structure of a small protein, avian pancreatic polypeptide. It was found that by making a large number of trials with sets of initially random phases a straightforward run of SAYTAN would yield phases for about 800 reflexions with a mean phase error less than 40°. From this point it was possible to phase extend to 2000 reflexions using SA YTAN and then, by recycling the information from E maps, eventually to obtain 2000 or even more reflexions with a mean phase error of order 28°. Final cycles of weighted Fourier syntheses led to maps which could easily be interpreted with models so leading to a complete elucidation of the structure. It was found that standard figures of merit, such as those used in MULTAN and SAYTAN and one based on negative quartets, were not effective in recognizing good sets of phases for large structures. The full exploitation of direct methods to solve protein structures awaits the discovery of a new figure of merit more effective in ranking trial phase sets.