Distributed structure determination at the JCSG

Abstract

The Joint Center for Structural Genomics (JCSG), one of four large-scale structure-determination centers funded by the US Protein Structure Initiative (PSI) through the National Institute for General Medical Sciences, has been operating an automated distributed structure-solution pipeline, Xsolve, for well over half a decade. During PSI-2, Xsolve solved, traced and partially refined 90% of the JCSG's nearly 770 MAD/SAD structures at an average resolution of about 2 Å without human intervention. Xsolve executes many well established publicly available crystallography software programs in parallel on a commodity Linux cluster, resulting in multiple traces for any given target. Additional software programs have been developed and integrated into Xsolve to further minimize human effort in structure refinement. Consensus-Modeler exploits complementarities in traces from Xsolve to compute a single optimal model for manual refinement. Xpleo is a powerful robotics-inspired algorithm to build missing fragments and qFit automatically identifies and fits alternate conformations.

Figure 1

Parallelization at the ‘program level’ in Xsolve. All outputs at each stage of Xsolve are distributed independently and in parallel to all programs at the next stage. Shown here are 14 combinations of software programs at the three stages in structure determination. autoSHARP includes model building with ARP/wARP and the resulting models are collected by ConsensusModeler. autoSHARP phases are input to Buccaneer and RESOLVE.

Figure 2

Superimposed traces represented as a graph. A docked trace is represented from left to right; traces resulting from different model-building protocols are represented on the vertical axis.

Figure 3

Fraction of residues from the sequence docked into the electron density for traces resulting from 36 reprocessed data sets. Depicted on the horizontal axis in the plane of the figure are the data sets, ranging in resolution from 1.3 to 3.0 Å. On the perpendicular horizontal axis are the processing strategies, with ARP/wARP traces in the foreground, RESOLVE traces in the middle and Buccaneer traces towards the back. The vertical axis represents the fraction of residues from the sequence docked into the electron density. Results are shown for the correct space group and number of molecules in the asymmetric unit and the most complete wavelength combination.

Figure 4

Number of times an indexing, phasing and model-building combination contributed the top trace for the 36 targets.

Figure 5

Percentage of improvement of the number of side chains docked to the electron density by the consensus model over the best input trace (blue line, left axis). The bars depict the number of input traces to ConsensusModeler. Only input traces with the correct space group and number of molecules in the asymmetric unit were considered. Wavelength combinations were binned, so that one input trace is reported for each program combination, similar to Fig. 3 ▶.

Figure 6

ConsensusModeler with two input models at 2.8 Å resolution. (a) The Buccaneer/SHARP/XDS model had 57% of the sequence docked into the model. (b) The RESOLVE/SHARP/XDS model had 39% of the sequence docked. (c) The consensus model resulted in 78% of the sequence docked.

Figure 7

ConsensusModeler-facilitated Xpleo. (a) The Buccaneer model had 92% of side chains docked. (b) The consensus model (cyan) calculated from seven input traces, with an 11-residue fragment added to the C-terminus and partially closing a 20-residue fragment from Met237 to Phe257. (c) Xpleo was able to fully close the remaining eight-residue gap from Asn249 to Phe257, resulting in a trace with 97% of residues docked. (d) The final refined model in green superimposed on the consensus model.