Efficient merging of data from multiple samples for determination of anomalous substructure.

Thomas C. Terwilliger,Janet L. Smith,David L. Akey

doi:10.1107/s2059798315021920

Abstract

Merging of data from multiple crystals has proven to be useful for determination of the anomalously scattering atomic substructure for crystals with weak anomalous scatterers (e.g. S and P) and/or poor diffraction. Strategies for merging data from many samples, which require assessment of sample isomorphism, rely on metrics of variability in unit-cell parameters, anomalous signal correlation and overall data similarity. Local scaling, anomalous signal optimization and data-set weighting, implemented in phenix.scale_and_merge, provide an efficient protocol for merging data from many samples. The protein NS1 was used in a series of trials with data collected from 28 samples for phasing by single-wavelength anomalous diffraction of the native S atoms. The local-scaling, anomalous-optimization protocol produced merged data sets with higher anomalous signal quality indicators than did standard global-scaling protocols. The local-scaled data were also more successful in substructure determination. Merged data quality was assessed for data sets where the multiplicity was reduced in either of two ways: by excluding data from individual crystals (to reduce errors owing to non-isomorphism) or by excluding the last-recorded segments of data from each crystal (to minimize the effects of radiation damage). The anomalous signal was equivalent at equivalent multiplicity for the two procedures, and structure-determination success correlated with anomalous signal metrics. The quality of the anomalous signal was strongly correlated with data multiplicity over a range of 12-fold to 150-fold multiplicity. For the NS1 data, the local-scaling and anomalous-optimization protocol handled sample non-isomorphism and radiation-induced decay equally well.

Highlights

The use of high-multiplicity Bijvoet data from multiple samples has been shown to be effective in the determination of anomalous substructure in difficult cases (Akey, Brown, Dutta et al, 2014; Liu et al, 2012)
While in principle increasing the multiplicity of merged data sets improves the data quality, complications arise from non-isomorphism when data from many samples are merged and from radiation damage when more data are collected from single crystals
Local scaling will better account for intensity differences that arise either from radiation damage or from errors in modelling sample absorption

Summary

Introduction

The use of high-multiplicity Bijvoet data from multiple samples has been shown to be effective in the determination of anomalous substructure in difficult cases (Akey, Brown, Dutta et al, 2014; Liu et al, 2012). These difficult cases include crystals with weak or few anomalous scatterers, poorly diffracting crystals and/or data collected far from the energy of peak absorption. We reported that in certain instances better substructure solutions resulted when ‘outlier’ data sets were excluded (Akey, Brown, Konwerski et al, 2014) This is confirmed when data are added (from best to worst) to create a merged data set (Terwilliger, Hung et al, 2016). While the signal-to-noise estimate of anomalous differences [h|ÁF|/Á(F)i] could be used, sigma values vary with the method used to estimate them and, especially for small differences and weak F, they may be an unreliable metric for evaluating different merging strategies

Data collection and processing

Findings

Summary