Premium
Database mining with adaptive fuzzy partition: Application to the prediction of pesticide toxicity on rats
Author(s) -
Pintore Marco,
Piclin Nadège,
Benfenati Emilio,
Gini Giuseppina,
Chrétien Jacques R.
Publication year - 2003
Publication title -
environmental toxicology and chemistry
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.1
H-Index - 171
eISSN - 1552-8618
pISSN - 0730-7268
DOI - 10.1002/etc.5620220505
Subject(s) - set (abstract data type) , fuzzy logic , partition (number theory) , linear subspace , class (philosophy) , data mining , computer science , test set , molecular descriptor , quantitative structure–activity relationship , chemical space , data set , toxicity , fuzzy set , artificial intelligence , machine learning , database , mathematics , chemistry , bioinformatics , biology , drug discovery , geometry , organic chemistry , combinatorics , programming language
A data set of 235 pesticide compounds, divided into three classes according to their toxicity toward rats, was analyzed by a fuzzy logic procedure called adaptive fuzzy partition (AFP). This method allows the establishment of molecular descriptor/chemical activity relationships by dynamically dividing the descriptor space into a set of fuzzily partitioned subspaces. A set of 153 molecular descriptors was analyzed, including topological, physicochemical, quantum mechanical, constitutional, and electronic parameters, and the most relevant descriptors were selected with the help of a procedure combining genetic algorithm concepts and a stepwise method. The ability of this AFP model to classify the three toxicity classes was validated after dividing the data set compounds into training and test sets, including 165 and 70 molecules, respectively. The experimental class was correctly predicted for 76% of the test‐set compounds. Furthermore, the most toxic class, particularly important for real applications of the toxicity models, was correctly predicted in 86% of cases. Finally, a comparison between the results obtained by AFP and those obtained by other classic classification techniques showed that AFP improved the predictive power of the proposed models.