Interpreting Microarray Expression Data Using Text Annotating the Genes

12 Microarray expression data is being generated by the gigabyte all over the world with 13 undoubted exponential increases to come. Annotated genomic data is also rapidly 14 pouring into public databases. Our goal is to develop automated ways of combining 15 these two sources of information to produce insight into the operation of cells under 16 various conditions. Our approach is to use machine-learning techniques to identify 17 characteristics of genes that are up-regulated or down-regulated in a particular micro18 array experiment. We seek models that are (a) accurate, (b) easy to interpret, and (c) 19 stable to small variations in the training data. This paper explores the effectiveness of 20 two standard machine-learning algorithms for this task: Na€ıve Bayes (based on prob21 ability) and PFOIL (based on building rules). Although we do not anticipate using our 22 learned models to predict expression levels of genes, we cast the task in a predictive 23 framework, and evaluate the quality of the models in terms of their predictive power on 24 genes held out from the training. The paper reports on experiments using actual E. coli 25 microarray data, discussing the strengths and weaknesses of the two algorithms and 26 demonstrating the trade-offs between accuracy, comprehensibility, and stability. 27 2002 Published by Elsevier Science Inc. 28 Information Sciences xxx (2002) xxx–xxx www.elsevier.com/locate/ins Corresponding author. E-mail addresses: [email protected] (M. Molla), [email protected] (P. Andreae), jeremy@ genome.wisc.edu (J.Glasner), [email protected] (F.Blattner), [email protected] (J. Shavlik). 1 On leave from Victoria University of Wellington, NZ. 0020-0255/02/$ see front matter 2002 Published by Elsevier Science Inc. PII: S0020 -0255 (02 )00216 -5 INS 6632 No. of Pages 14, DTD=4.3.1 12 August 2002 Type SPS, Chennai ARTICLE IN PRESS