Naslov
Improved structure-toxicity relationships for aquatic toxicity of environmental pollutants

Autori
Lučić, Bono ; Amić, Dragan ; Novič, Marjana ; Nadramija, Damir ; Bašic, Ivan

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
MATH/CHEM/COMP 2004, Book of abstracts / Graovac, Ante - Zagreb, 2004

Skup
MATH/CHEM/COMP 2004

Mjesto i datum
Dubrovnik, Hrvatska, 21.06.2004. - 26.06.2004

Vrsta sudjelovanja
Poster

Vrsta recenzije
Nije recenziran

Sažetak
Numerous organic chemicals can be environmental pollutants, and due to this fact, many studies were directed towards the understanding of relationships between a structure and toxicity of a compound. Structure-toxicity models are strongly dependent on the class of molecules for which models are obtained. Classification of molecules is defined by the mechanism of the toxic action of molecules, and this piece of information can be obtained experimentally, or predicted by developed algorithms (O. Ivanciuc, Internet Electron. J. Mol. Des. 2002, 1, 157 ; A. O. Aptula et al. Quant. Struct.-Act. Relat., 2002, 21, 12). We will test the degree of improvement of existing models for predicting toxicity of molecules by using structural descriptors computed by the Dragon program (R. Todeschini ; V. Consonni, http://www.disat.unimib.it/chm/), and by including information on the mechanism of toxic action. Analysis was based on 293 organic molecules for which experimental aqueous toxicity on Poecilia reticulata were collected from the literature (A. R. Katritzky et al. J. Chem. Inf. Comput. Sci. 2001, 41, 1162). Starting from structures encoded as the SMILES string, molecular structures were generated by the CORINA 3D structure generator (www2.chemie.uni-erlangen.de/software/corina/). Molecules were characterized by more than 800 molecular descriptors that were filtered in order to remove highly inter-correlated descriptors. Final models were selected by the CROMRsel program for efficient selection of a small sub-set of the most important descriptors into the multiregression models. We obtain simple multivariate form of models containing 2-8 optimized parameters (i.e. 1-7 of descriptors) with better statistical performance than the published models developed on the same set of molecules.

Izvorni jezik
Engleski

Znanstvena područja
Kemija

POVEZANOST RADA