Structure Quality

A COMPREHENSIVE STUDY of 14,518 crystal structures in the Protein Data Bank (PDB) reports what it calls "a striking result"—high-prestige journals tend to publish structures with more errors.

The study finds that "the worst offenders" are the most highly competitive journals, such as Cell, Science, Molecular Cell, and Nature. Protein structure specialists have long suspected this, but the new report quantifies the effect in a systematic manner.

The investigation also finds that structures obtained by structural genomics centers are more accurate than those generated by other labs. Structural genomics groups determine structures by using high-throughput, automated methods, whereas other structure specialists use a slower, more hands-on approach. The genomics techniques have been criticized as potentially more error-prone, but the new report refutes that charge.

Grad student Eric N. Brown and associate professor of biochemistry Subramanian Ramaswamy of the University of Iowa, Iowa City, conducted the study (Acta Crystallogr. Sect. D 2007, 63, 941). They conclude that "the rush to publish high-impact work" may help explain "the proliferation of poor-quality structures" documented by the study.

Katrina Kelner, deputy editor for life sciences at Science, says high-impact journals tend to publish reports on more complex proteins, "and it is not surprising that these results at the forefront of the field are less well-defined than other, more routine structures." The new analysis corrected for structure size and novelty but not explicitly for complexity, such as the existence of multiple proteins and nucleic acids in structures, Kelner notes. But she believes the work makes a positive contribution to the field.

Alexander Wlodawer, chief of macromolecular crystallography at the National Cancer Institute, Frederick, Md., points out that a series of structure papers that had been cited by hundreds of other papers were retracted recently, "because they were completely wrong." He suggests that automated analyses of structure quality generated routinely by PDB be made available to reviewers of structure papers. "Then reviewers would at least know what to think about the quality of structures," he says. "Anything that can be corrected and could be done better should be done better."

Professor Richard N. Armstrong of Vanderbilt University School of Medicine is the editor of the ACS journal Biochemistry, which fared among the best in the analysis. He says, "One question any researcher should ask before rushing off a competitive structure report to any journal is: 'Do I want to be first or best?' If the answer is 'first,' then it is time for some serious reflection as to what science is about."