Highlighting a recent publication about GO
Interaction networks derived from high-throughput (HTP) methods, such as proteome-wide purification of protein complexes using affinity-tagged proteins, have become important resources for both gene function prediction and studies of complex biological networks, but the specificity of these networks has been difficult to assess. Data published in a recent paper from Mike Tyers' lab (University of Toronto) and incorporated within their BioGRID resource (http://www.thebiogrid.org) compares the results of HTP datasets versus that of individual experiments described in the literature. The paper, Reguly et al., The Journal of Biology, 2006;5(4):11, describes a comprehensive dataset of genetic and physical interactions for the budding yeast Saccharomyces cerevisiae. Using GO Biological Process annotations curated from the primary literature by the Saccharomyces Genome Database (SGD; http:// www.yeastgenome.org/), the authors analyzed the extent of shared GO annotations between interaction pairs defined in the literature-curated (LC) dataset versus those defined in HTP datasets. The results indicate that inclusion of the LC dataset within their analysis significantly improves gene function prediction. The LC dataset assembled by Reguly et al. thus provides not only a valuable new resource for annotating a gene's role in biology, but an important benchmark by which the accuracy of HTP datasets can be measured. The complete LC dataset is available from BioGRID and mirrored at SGD.