MP6 Quantitative Structure Activity Relationships (QSARs)
Monday, 14 November 2005: 8:00 AM - 5:30 PM in Exhibit Hall

(BAJ-1117-828648) The significance of transformation in data mining for quantitative structure-toxicity relationships.

Baja, M. Sc., Emmanuel¹, Patungan, Ph.D, Welfredo ², Pizana, Ph.D., Romulo³, Quibuyen, Ph.D., Titos⁴, Hermosilla, Ph.D., Augusto³, ¹ Department of Environmental Medicine, New York University School of Medicine, Tuxedo, NY, USA² School of Statistics, University of the Philippines Diliman, Quezon City, Philippines³ Department of Mathematics, University of the Philippines Diliman, Quezon City, Philippines⁴ Institute of Chemistry, University of the Philippines Diliman, Quezon City, Philippines

ABSTRACT- A Quantitative Structure-Toxicity Relationship (QSTR) approach based on Multiple Linear Regression (MLR) was used to check if transformations on the response function and descriptors would give a better response-factor relationship. Twenty linear models were used in the study using data sets from forty three carbamate insecticide derivatives. LD₅₀ values of the carbamates were used as the response variable, while the descriptors used were solvent-accessible surface area, volume, mass, van der Waals surface area, partial charges, refractivity, polarizability and octanol-water partition coefficient. The values of the physicochemical and structural descriptors were calculated for the functional groups (R₁ and R₂) attached to the parent compound of the carbamates. Coefficient of determination R², estimate of error variance s², and Fisher test (F-test) p-value were used as criteria for the model selection process. Multiple linear regression analysis showed that the 1/Y transformation of the original data yielded a number of benefits. The transformation helped to normalize the error distribution, stabilized the variance and simplify the response function and descriptors by linearizing a nonlinear dependent-independent relationship.

Key words: quantitative structure-toxicity relationship (QSTR), multiple linear regression (MLR), transformation, carbamates, and LD₅₀