LOW RESOLUTION AB-INITIO PHASING WITH MONTE CARLO AND CLUSTERIZATION TECHNIQUES. V.Y. Lunin, Institute of Mathematical Problems of Biology, Russian Academy of Sciences, Pushchino, Moscow Region, 142292, Russia

A Monte Carlo type approach has been developed for low-resolution ab-initio phasing. It is based on the generation of a large amount of possible phase sets followed by an enrichment procedure which rejects non-admissible sets in accordance with some specified selection criteria. Two approaches to phase sets generation were tried: direct generating of phase values (Lunin, Acta Cryst., D49, 90-99, 1993) and the recently developed Few Atom Models method (Lunin et al., Acta Cryst., D51, 896-903, 1995) in which low resolution phase sets are approximated using structure factors calculated from pseudo-atomic models.

The various selection criteria (suggested by different authors), such as magnitude correlation, electron density map (e.d.m.) histograms, e.d.m. connectivity and local density variation, maximum likelihood estimates of phase errors etc. are not strongly discriminative when applied in the low resolution range. Furthermore, attempts at local refinement without special precautions fits the criteria without necessarily improving the e.d.m.. (Lunin&Skovoroda, Acta Cryst., A51, 880-887, 1995). To overcome this difficulty a cluster analysis procedure was applied to split the enriched collection of phase sets into a small number of clusters, each representing a possible solution. This procedure as well as the averaging of variants inside a cluster require preliminary maps alignment (Lunin & Lunina, Acta Cryst., in print, 1996).Some additional criteria, e.g. high density at dyads, may be used then to reject wrong clusters.

The low resolution phase information obtained may be used to construct non-trivial prior coordinate probability distributions and to modify classical direct methods approaches on the base of explicit representation of saddle-point based approximations for the whole set of structure factors (Lunin et al., to be published).