Despite the diversity of topics and speakers, some common threads emerged at the joint structural biology meetings in Keystone this past week. First, structural genomics clearly has hit its stride. The US Protein Structure Initiative deposited some 1,300 structures in the Protein Data Bank between 2000 and 2005, RIKEN added 1,347 of its own between 2002 and 2005, and the Structural Genomics Consortium added another 180 in the past 18 months or so. That?s nearly 3,000 structures in five years. The PDB required 20 years to first achieve that number, only surpassing it in 1996.

There remain many proteins that are tough to work with, either because they represent traditionally difficult classes (like membrane proteins), or because they cannot be made in soluble form. New techniques were presented to deal with these problems, but it?s clear that as one bottleneck opens, another closes. Thus the cost of the typical structure is tough to measure. According to PSI head John Norvell the four best structural genomics centers were producing structures at $87,000 per in 2005. But Zygmunt Derewenda pointed out that the gene-to-structure cost (as opposed to the crystal-to-structure cost) remains high, because the crystallization process remains so temperamental.

Now all eyes are on PSI Phase 2 [press release ], in which four production centers will work to generate between 3,000 and 4,000 new structures while six specialized centers will work to further improve the efficiency of structural work for traditionally difficult protein classes, including eukaryotic and membrane proteins. Determining exactly which proteins to target evidently is a touchy subject, however, given the number of times it was addressed. Said Norvell, there are as many opinions on target list composition at the meeting as there were participants, and possibly more.

Most structural genomics work to date has involved prokaryotic homologs of eukaryotic proteins, because eukaryotic proteins are much tougher to work with. William Nierman has an idea why that might be: Estimating that some two-thirds of gene models in eukaryotic databases contain errors, Nierman (of the Institute for Genomic Research) suggested that structural biologists' relatively poor ability to crystallize eukaryotic proteins might be due in part to poor genome annotation. Clearly, this is an area that must be addressed if eukaryotic structural genomics is ever to take off.

In a corollary to this idea, Tom Blundell of the University of Cambridge suggested that the PDB might also suffer harbor unacceptable accuracy. Though several structure solutions are sometimes possible from a given crystal, biologists often post only the most likely fit, he noted. The database should therefore be modified to accept an ensemble of the best crystal conformers to improve its accuracy, much as is done with NMR structures. In the ensuing Q&A session, a representative of the PDB noted that the technology used to handle NMR data could be adapted to x-ray structures, so this modification could be forthcoming.

Finally, there was some discussion of the need for standardization and benchmarking of structural biology techniques. Par Nordlund, for instance, of the Structural Genomics Consortium, Stockholm, presented hard data to compare new techniques for rapidly identifying soluble eukaryotic proteins in E. coli against existing protocols. It?s hard to argue with his point: only through careful analysis can researchers effectively design and optimize their workflows.