A knowledge-driven probabilistic framework for the prediction of protein-protein interaction networks

Fiona Browne, Haiying Wang*, Huiru Zheng, Francisco Azuaje

*Corresponding author for this work

    Research output: Contribution to journalArticleResearchpeer-review

    15 Citations (Scopus)


    This study applied a knowledge-driven data integration framework for the inference of protein-protein interactions (PPI). Evidence from diverse genomic features is integrated using a knowledge-driven Bayesian network (KD-BN). Receiver operating characteristic (ROC) curves may not be the optimal assessment method to evaluate a classifier's performance in PPI prediction as the majority of the area under the curve (AUC) may not represent biologically meaningful results. It may be of benefit to interpret the AUC of a partial ROC curve whereby biologically interesting results are represented. Therefore, the novel application of the assessment method referred to as the partial ROC has been employed in this study to assess predictive performance of PPI predictions along with calculating the True positive/false positive rate and true positive/positive rate. By incorporating domain knowledge into the construction of the KD-BN, we demonstrate improvement in predictive performance compared with previous studies based upon the Naive Bayesian approach.

    Original languageEnglish
    Pages (from-to)306-317
    Number of pages12
    JournalComputers in Biology and Medicine
    Issue number3
    Publication statusPublished - Mar 2010


    • "Omic" datasets
    • Computational systems biology
    • Functional genomics
    • Machine and statistical learning
    • Protein-protein interaction networks


    Dive into the research topics of 'A knowledge-driven probabilistic framework for the prediction of protein-protein interaction networks'. Together they form a unique fingerprint.

    Cite this