Mining patterns in disease classification forests

    Authors

    H. Y. Hu

    Comments

    Authors: contact us about adding a copy of your work at STARS@ucf.edu

    Abbreviated Journal Title

    J. Biomed. Inform.

    Keywords

    Biological pathway; Pattern mining; Random forests; Disease phenotype; LUNG-CANCER CELLS; C PKC ISOFORMS; BREAST-CANCER; EXPRESSION PROFILES; MAMMARY-GLAND; PATHWAY; ALPHA; ACTIVATION; REGRESSION; SURVIVAL; Computer Science, Interdisciplinary Applications; Medical Informatics

    Abstract

    Multiple biological pathways often work together to determine a given disease phenotype. Understanding what these pathways are and how they cooperate in disease-relevant biological processes is critical to our understanding of diseases. Using microarray gene expression data, researchers have developed several methods to rank pathways by their disease relevance. However, the exact set of pathways involved and how they are involved under given disease conditions remain unclear. In this paper, we propose a novel method to first select a robust set of pathways that together best classify a given disease, and then investigate how genes in these pathways interact to determine the phenotype. By applying our method to several disease related microarray gene expression datasets, we detected many disease-relevant interaction patterns supported by evidence from the literature. Our algorithm also achieves higher accuracy in terms of identification of a robust set of disease-relevant pathways when compared with alternative strategies. Published by Elsevier Inc.

    Journal Title

    Journal of Biomedical Informatics

    Volume

    43

    Issue/Number

    5

    Publication Date

    1-1-2010

    Document Type

    Article

    Language

    English

    First Page

    820

    Last Page

    827

    WOS Identifier

    WOS:000281927200019

    ISSN

    1532-0464

    Share

    COinS