Pesticide Persistence in the Environment - Collected Data and Structure-Based Analysis

Sokratis ALIKHANIDI and Yoshimasa TAKAHASHI*

Department of Knowledge-based Information Engineering, Toyohashi University of Technology,
Tempaku-cho, Toyohashi 441-8580, Japan

(Received: October 23, 2003; Accepted for publication: January 23, 2004; Published on Web: March 22, 2004)

A data set of 420 pesticide persistences in the environment was collected as field half-life (HL) using several on-line databases. Due to the fuzziness of observed values, the compounds were grouped into three major categories: class 1 when a pesticide has HL <= 30 days; class 2, if 30 < HL <= 100 days; and class 3, if HL > 100 days. The Quantitative Structure-Biodegradation Relationship (QSBR) analysis was worked out on the training set of 315 pesticides. Thirty one topological substructural descriptors were used and the decision tree approach was employed for the modeling. Estimation results were as follows: for the train set, 5 compounds of two-unity (class 1/class 3) misclassification, 38 compounds of unity (class 1/class 2 and class 2/class 3) misclassification, and 272 compounds (86.3%) were correctly classified; for the test set, there were 3, 20, and 82 compounds (78.1%) respectively. The computer expert system EKeeper was developed on the basis of the QSBR model.

Keywords: QSAR, QSBR, Data mining, Estimation of biodegradation, Expert system, Pesticide persistence

Abstract in Japanese

Text in English

PDF file(367kB)