GNN - Genome News Network  
  Home | About | Topics
An automated system for classifying related proteins by structure
Edward R. Winstead

Scientists have created an automated system for assigning evolutionary relationships to proteins based on their structures. The system is capable of recognizing similarities in folding patterns between protein pairs and can partition family trees of structures into more defined groups, or 'superfamilies.' Until now, scientists, not computer algorithms, have classified protein structures.

Liisa Holm and Sabine Dietmann, of the Structural Genomics Group, EMBL-EBI, in Cambridge, UK, developed the method. They tested the system against a classification done by human experts and found a 92 percent reliability rate. "The method is fully automated, allowing fast, self-consistent and complete classification of large numbers of protein structures," they write in Nature Structural Biology.

Classifying proteins by structure is potentially useful to biologists because proteins with similar structures may interact with each other, or may share similar functions. But structural elements do not tell the whole story. Different members of a single superfamily can be involved in unrelated biochemical activities.

The manual approach will be important in overseeing the classification of homologous proteins, writes Olivier Lichtarge, of Baylor College of Medicine, in Houston, in an associated News and Views piece. "At the same time," he adds, "the automation of this process is critical to cope with a massive inflow of data and to shed light on the biases and inconsistencies that humans will inevitably introduce."

. . .

Dietmann, S. & Holm, L. Identification of homology in protein structure classification. Nat Struct Biol 8, 953-957 (November 2001).
Lichtarge, O. Getting past appearances: the many-fold consequences of remote homology. Nat Struct Biol 8, 918-920 (November 2001).

Back to GNN Home Page