https://doi.org/10.15255/KUI.2003.030
Published: Kem. Ind. 53 (7-8) (2004) 323–331
Paper reference number: KUI-30/2003
Paper type: Original scientific paper
Download paper:  PDF

Application of Artificial Neural Networks to the QSPR Study – Automated Classification of Endocrine Disrupting Chemicals

M. Novič and A. Roncaglioni

Abstract

The European Union Commission delivered a list of 553 chemicals that were inspected for the scientific evidence of their endocrine disruption activity. The source of information, i.e. the studies collected in the report refer to experiments made during the decades, evaluating several species and a great variety of effects, which reflects in non-homogeneity of the data. The classification of potential endocrine disrupters (EDs), according to the literature evidence of their functioning, was proposed by the Commission. The endocrine disruption categories given in the EU Commission report are the following: (i) certainly active as endocrine disrupters, (ii) potentially active, (iii) less probable active – lacking evidence, and (iv) certainty non-active. The research of the methodology to find an automated predictive model, yielding the ED categories, is presented. Clustering and classification techniques were employed to solve the problem. From the list of 553 chemicals a dataset of 106 molecules with the defined chemical structure and ED class were extracted. Molecular structures were represented by 3D atomic coordinates calculated with the AM1 or PM3 semi-empirical method for all 106 chemicals. From 3D coordinates an extensive set of molecular descriptors was calculated. The classification model based on counterpropagation neural network (CP NN) was prepared and evaluated. The method of determining the thresholds necessary to co

This work is licensed under a Creative Commons Attribution 4.0 International License

Keywords

QSPR study, endocrine disrupting chemicals