PARENT SESSION

1F - QSAR
Poster Hall
8:30 AM - Tuesday, 29 April 2003
Chair: Schüürmann, G.¹, ¹
Co-chair: Verhaar, H.J.M.², Cronin, M.³, ^{2 3}

(TUP/44) Impact of Variable Selection on QSARs for the Toxicity of Phenols to Tetrahymena pyriformis.

Netzeva, Tatiana¹, Aptula, Aynur ², Novič, Mariana³, Schultz, Terry⁴, Schüürmann, Gerrit², Tropsha, Alexander⁵, Xiao, Y⁵, Cronin, Mark¹, ¹ Liverpool John Moores University, Liverpool, Merseyside, England² UFZ Centre for Environmental Research, Leipzig, Leipzig-Halle, Germany³ National Institute of Chemistry, Ljubljana, Ljubljana, Slovenia⁴ The University of Tennessee, Knoxville, TN, USA⁵ University of North Carolina, Chapel Hill, NC, USA

ABSTRACT- Quantitative structure-activity relationships (QSARs) are being applied to an increasingly diverse number of toxicological endpoints. The diversity of endpoints being modelled will inevitably result in more complexity in the mechanisms of toxic action, hence requiring greater complexity in the modelling process. One method to model greater mechanistic complexity is the use of more physico-chemical and/ or structural descriptors. However, the use of a greater numbers of descriptors will undoubtedly require the use of some form of selection technique to determine those best suited to modelling the endpoint. The aim of this study was to determine the impact of variable selection procedures on different models for the acute toxicity of phenols to Tetrahymena pyriformis. To meet this goal a large number of physico-chemical, topological, electrotopological and electronic descriptors were calculated for more than 200 phenols. A variety of multiple linear regression (MLR), partial least squares, artificial neural network and K-nearest neighbour (k-NN) approaches were utilised for variable selection and model development. The number of selected descriptors varied from 2 (MLR) to 25 (k-NN). All the models were highly predictive but the transparency and interpretability of the multi-descriptor models decreased with the increasing number of descriptors in the model. The different approaches to descriptor selection with techniques influenced the final outcome of the models. [This work was supported in part by the European Union IMAGETOX Research Training Network (HPRN-CT- 1999-00015)].

Key words: multivariate statistics, QSAR, variable selection, phenols