A NEW FIGURE OF MERIT FOR PROTEIN PHASE SETS DISCRIMINATION
Anatoly Mishnev
21 Aizkraukles St., LV 1006, Riga, Latvia, E-mail: mishnevs@osi.lanet.lv
Keywords: direct methods, protein phasing,
phase discrimination, autocorrelation function
The success of multisolution direct methods (DM) when applied to protein structures is hardly dependent on a suitable figure of merit (FOM) able to reliably identify the correct solution. In recent years it has been shown that conventional FOMs are not effective in picking out the best solutions from a large number of phase sets. Here we describe a new ACFFOM based on the general properties of the autocorrelation function (ACF) for a single unit cell and its relation to the Patterson function. It has been shown that the ACF is dependent on the phases of the structure factors (SF) [1]. When phase estimates are available from DM the ACF can be calculated for each phase set and compared with the Patterson function. For intramolecular vectors correct phases will produce the ACF nearly coincident with the Patterson. This approach has a similarity with molecular replacement method where the "known" structure is replaced by the electron density calculated using DM phases. To distinguish between good and bad phase sets we consider the values of the ACF A(u) at Patterson maxima. It is supposed that ACF for a good phase set should have maxima at the same positions as Patterson function. These considerations lead to the following FOM
ACFFOM = Si A2(ui) ,
where the sum is over the grid points corresponding to P(u) maxima and which assumes the highest value for the true phases..
The procedure of ACFFOM calculation is as follows:
The SAYTAN program [2] was used to derive 200 sets of
phases for a small 300 non-hydrogen atoms protein APP [3]. Three
of the phase sets had unweighted mean phase errors in the range
of 38 - 42o. The ACFFOM was successfully
applied to discriminate these good solutions.