Bethany J Wolf1, Elizabeth G Hill, Elizabeth H Slate. 1. Division of Biostatistics and Epidemiology, Medical University of South Carolina, 135 Cannon St., Charleston, SC, USA. wolfb@musc.edu
Abstract
MOTIVATION: Highly sensitive and specific screening tools may reduce disease -related mortality by enabling physicians to diagnose diseases in asymptomatic patients or at-risk individuals. Diagnostic tests based on multiple biomarkers may achieve the needed sensitivity and specificity to realize this clinical gain. RESULTS: Logic regression, a multivariable regression method predicting an outcome using logical combinations of binary predictors, yields interpretable models of the complex interactions in biologic systems. However, its performance degrades in noisy data. We extend logic regression for classification to an ensemble of logic trees (Logic Forest, LF). We conduct simulation studies comparing the ability of logic regression and LF to identify variable interactions predictive of disease status. Our findings indicate LF is superior to logic regression for identifying important predictors. We apply our method to single nucleotide polymorphism data to determine associations of genetic and health factors with periodontal disease. AVAILABILITY: LF code is publicly available on CRAN, http://cran.r-project.org/.
MOTIVATION: Highly sensitive and specific screening tools may reduce disease -related mortality by enabling physicians to diagnose diseases in asymptomatic patients or at-risk individuals. Diagnostic tests based on multiple biomarkers may achieve the needed sensitivity and specificity to realize this clinical gain. RESULTS: Logic regression, a multivariable regression method predicting an outcome using logical combinations of binary predictors, yields interpretable models of the complex interactions in biologic systems. However, its performance degrades in noisy data. We extend logic regression for classification to an ensemble of logic trees (Logic Forest, LF). We conduct simulation studies comparing the ability of logic regression and LF to identify variable interactions predictive of disease status. Our findings indicate LF is superior to logic regression for identifying important predictors. We apply our method to single nucleotide polymorphism data to determine associations of genetic and health factors with periodontal disease. AVAILABILITY: LF code is publicly available on CRAN, http://cran.r-project.org/.
Authors: Charles Kooperberg; Joshua C Bis; Kristin D Marciante; Susan R Heckbert; Thomas Lumley; Bruce M Psaty Journal: Am J Epidemiol Date: 2006-11-02 Impact factor: 4.897
Authors: K S Kornman; A Crane; H Y Wang; F S di Giovine; M G Newman; F W Pirk; T G Wilson; F L Higginbottom; G W Duff Journal: J Clin Periodontol Date: 1997-01 Impact factor: 8.728
Authors: Claudia Maria Astolfi; André Luís Shinohara; Rodrigo Augusto da Silva; Maria Cristina Leme Godoy Santos; Sergio Roberto Peres Line; Ana Paula de Souza Journal: J Clin Periodontol Date: 2006-08-07 Impact factor: 8.728
Authors: Bethany J Wolf; Elizabeth G Hill; Elizabeth H Slate; Carola A Neumann; Emily Kistner-Griffin Journal: PLoS One Date: 2012-11-08 Impact factor: 3.240
Authors: Bethany J Wolf; Paula S Ramos; J Madison Hyer; Viswanathan Ramakrishnan; Gary S Gilkeson; Gary Hardiman; Paul J Nietert; Diane L Kamen Journal: Genes (Basel) Date: 2018-10-15 Impact factor: 4.096
Authors: Christophe Lalanne; Bruno Falissard; Bernard Golse; Laurence Vaivre-Douret Journal: BMC Med Res Methodol Date: 2012-07-26 Impact factor: 4.615