Publication was possible with the founding of PIFI. “
“New technologies such as DNA microarray
and next-generation sequencer have allowed researchers to learn Akt inhibitor biological phenomena in genome or transcriptome levels. Especially in toxicology, these new technologies have led to a new subdiscipline, termed toxicogenomics. Toxicogenomics is concerned with the identification of potential human and environment toxicants, and their putative mechanisms of action, through the use of genomics resources [1]. For example, by evaluating and characterizing differential gene expressions, in humans or animals, after exposure to drugs, it is possible to use complex expression patterns to predict toxicological outcomes and to identify mechanisms involved with or related to the toxic event [2]. Traditionally, to construct a such predictive classifier, techniques in machine learning such as k-nearest neighbors, linear discriminant analysis (LDA) and support vector machine (SVM) have been mostly used [3]. However, building a classifier that is accurate and understandable at the same time is not necessarily an easy task. For example, while SVM achieves high classification accuracy, resulting classifiers are hard to interpret
as variables are transformed nonlinearly into a feature space, and hence difficult to use in order to extract relevant biological knowledge from it [4]. Very often, predictive accuracy, understandability, and computational demands need to be traded off against one another, because algorithms often compromise Selleckchem Epacadostat one to gain performance in the other [5]. In this study, we applied the Classification Based on Association (CBA) algorithm to toxicogenomic Tryptophan synthase data in an aim to build a classifier that is accurate and understandable at the same time. We compared its predictive performances and interpretability of generated classifiers with those of LDA, which is considered to be one of the most standard classification methods and have a good balance between accuracy and interpretability. CBA is one of the Class Association
Rule (CAR) mining algorithms, which integrate association rule mining (finding all the rules existing in the database that satisfy some constraints) and classification rule mining (discovering a small set of rules in the database that forms an accurate classifier) by focusing on mining a special subset of association rules, called class association rules (CARs) [6]. One of the advantages of CAR mining algorithms over conventional methods (especially SVM) is its interpretability, because classifiers are generated as a set of simple rules without much sacrifice of accuracy [7]. Another advantage is that CAR mining algorithms can be applied not only to linearly separable cases, but also to linearly inseparable cases, where LDA or other linear classification methods are not applicable [8].
Related posts: