Protein Domain Boundary Predictions: A Structural Biology Perspective

Kirillova, Svetlana; Kumar, Suresh; Carugo, Oliviero

In the present manuscript we have analyzed the reliability of the predictions that were made in the CASP7 experiment and that are publicly available. It was found that most of the bioinformatics tools are able to determine if a protein is made by a single domain or if it contains more than one domain, despite a similar reliability is reached by considering only the sequence length, a much simpler strategy. Using a standard and well known statistical test, we showed that most of the predictions that can be done are not impressively better than pseudo-random predictions. It was also observed that although the reliability of the prediction methods seems to be insufficient to make them routine tools in experimental structural biology, their performance can be extremely good. When the domain is correctly identified, its boundaries are very close, within one or two residues, to the experimental ones. In conclusion, these bioinformatics applications are not yet sufficiently accurate to be used as routine tools in experimental structural biology. It is rather probable that the use of more than a single prediction method by a sort of consensus approach might improve the reliability of the predictions. Although these bioinformatics tools are still immature, progress can be expected in the future. This work was supported by the Austrian GEN-AU project BIN-II. Björn Sjöblom and Kristina Djinovic are acknowledged for helpful discussions. Financial support by Putta None is also acknowledged. One reviewer is acknowledged for a series of references that deserved citation.