Estimating conditional probabilities for the detection of unfavorable copy number alterations in a targeted therapy

Fang Han Hsu, Edward R. Dougherty, Yidong Chen, Erchin Serpedin

    Research output: Contribution to journalArticlepeer-review

    1 Scopus citations


    Emerging targeted therapies have shown benefits such as less toxicity and higher effectiveness in specific types of cancer treatment; however, the accessibility of these advantages may rely on correct identification of suitable patients, which remains highly immature. We assume that copy number profiles, being accessible genomic data via microarray techniques, can provide useful information regarding drug response and shed light on personalized therapy. Based on the mechanism of action (MOA) of trastuzumab in the HER2 signaling pathway, a Bayesian network model in which copy number alterations (CNAs) serve as latent parents modifying signal transduction is applied. Two model parameters M-score and R -value which stand for the qualitative and quantitative effects of CNAs on drug effectiveness and are functions of conditional probabilities (CPs), are defined. An expectation-maximization (EM) algorithm is developed for estimating CPs, M-scores, and R-values from continuous measures, such as microarray data. We show through simulations that the EM algorithm can outperform classical threshold-based methods in the estimation of CPs and thereby provide improved performance for the detection of unfavorable CNAs. Several candidates of unfavorable CNAs to the trastuzumab therapy in breast cancer are provided in a real data example.

    Original languageEnglish (US)
    Article number6524018
    Pages (from-to)2933-2942
    Number of pages10
    JournalIEEE Transactions on Biomedical Engineering
    Issue number10
    StatePublished - 2013


    • Bayesian network
    • copy number
    • drug response
    • expectation-maximization algorithm
    • gene expression

    ASJC Scopus subject areas

    • Biomedical Engineering


    Dive into the research topics of 'Estimating conditional probabilities for the detection of unfavorable copy number alterations in a targeted therapy'. Together they form a unique fingerprint.

    Cite this