1. RESULT and DISCUSSION In order to find a relation between the three rate reaction constant (k OH, k NO3 and k O3 ) and the structural features of chemicals, a wide set of theoretical molecular descriptors were calculated by the software DRAGON[4], the calculated descriptors being able to catch and represent the different aspects of the molecular structures, viewing the molecules is one-dimensional, two-dimensional and three-dimensional ways. An advantage of the exclusive use of the theoretical descriptors is that they are free of the uncertainty of experimental measurements ( apart from “noise” introduced in the model by the training data set). For a stonger evaluation of model applicability for prediction on new chemicals, the external validation (verified by Q 2 ext ) of all models is also recommended [10,11] and was here performed. Experimental design provides a strategy for selecting thr most dissimilar molecular structures in a data set, talking into account the complete structural information obtained from all the used molecular descriptors and also the response value. Therefore it has guaranteed that the chemical composition of the training and external validations set have well balanced structural diversity and are also rapresentative of the entire range of the response. MATERIALS and METHODS QSAR PREDICTIONS OF ORGANIC COMPOUND TROPOSHERIC DEGRADATIONS Gramatica Paola*-Pilutti Pamela-Papa Ester-Battaini Francesca-Pozzoli Luca Dep.Struct.Funct.Biol. QSAR Research Unit - University of Insubria ( Varese - Italy ) Web: http://fisio.dipbsf.uninsubria.it/qsar/ e-mail: paola.gramatica@uninsubria.it ABSTRACT The reactions of chemicals with OH, NO 3 radicals and ozone are the principal degradation processes in the troposphere, thus an upper limit of chemical atmospheric persistence is assessed by determining the rate costants of such reactions. Our goal is to develop QSAR models that make it possible to rapidly predict the atmospheric degradability of organic chemicals from a simple description of the molecular structure. The molecular descriptors used are 1D-, 2D- and 3D-descriptors (constitutional, topological, WHIM, GETAWAY, quantum chemical and others). The best descriptor subsets for each modelling were selected using Genetic Algorithm-Variable Subset Selection strategy (GA-VSS) and model calculations were performed by Ordinary Least Squares regression (OLS). In order to render the predictions for new compounds more reliable, the models were validated using leave-one-out, leave-more-out (50%) procedures, external validation and the scrambling of responses; the reliability of the predictions was always checked by the leverage approach. All the obtained models were satisfactory at different levels (Q 2 =83-90%). A PCA model based on the three principal degradation process has been proposed to evaluate the overall atmospheric persistence of chemicals; the PC1 score obtained is proposed as an At mospheric P ersistence In dex ( ATPIN ) and is also modelled by theoretical molecular descriptors. This model can be used as an evaluative model for the screening and ranking of chemicals according to their atmospheric persistence just starting from their chemical structure. INTRODUCTION The troposphere is the principal recipient of volatile organic chemical (VOC) both of anthropogenic and biogenic origin. The persistence of these chemicals is one of the most relevant factors for the evalutation of their fate and possible negative effects, in the enviromental risk assessment. The tropospheric lifetime of most of the organic compounds from terrestrial emissions are controlled by the degradation reaction with the OH radical and ozone during the day and NO 3 radical at night. New predictive QSAR models for oxidation rate costant ( k OH, k NO3 and k O3 ), based on different theoretical molecular descriptors, selected by Genetic Algorithm as variable subset selection, are proposed. To make a reliable risk assesment of a large group of chemicals, model prediction capability (Q 2 ) is considered of primary importance and was evaluated using internal (leave-one-out and leave -more-out) and external validation procedure. The splitting of the original data set has been obtained by the Experimental Design. EXPERIMENTAL DATA Reaction rate costants with OH radical, NO 3 radical and Ozone for a total of 504 different organic compounds at 298 K were taken from Atkinson[1,2,3]. Experimental data are reported in cm 3 ·s -1 ·mol -1 and transformed in logarithmic units, then moltiplied by –1 in order obtain positive values. MOLECULAR DESCRIPTORS The molecular descriptors were calculated by the software DRAGON of Todeschini et al[4]. A total of 1150 molecular descriptor of differents kinds were used to describe compound chemical diversity. The descriptor typology is: In addition, four quantum-chemical descriptors (HOMO, LUMO, (HOMO-LUMO)GAP, energies and ionization potential Eiv),calculated by MOPAC (PM3 method)[5] were always added as molecular descriptors. 0D-costitutional 1D-functional groups 1D-atom centred fragments 1D-empirical 2D-topological 2D-BCUTs 2D-Galvez Indices 2D- walk counts 2D-varius autocorrelations from the molecular graph 3D-Randic molecular profiles 3D-geometrical 3D-WHIMs 3D-GETAWAYs CHEMOMETRIC METHODS Multiple Linear Regression analysis and variable selection were performed by the software MOBY-DIGS of Todeschini et al.[6], using the Ordinary Least Squares regression (OLS) method and GA-VSS[9]. External validation were performed on two validation sets obtained with the splitting at 50% and 75% of the original data set by Experimental Design procedure, applying the software DOLPHIN of Todeschini et al[7]. Tools of regression diagnostics as residual plots and Williams plots were used to check the quality of the best models and define their applicability regarding to the chemical domain, using the chemometric package SCAN[8]. RMS (residual mean squares) are also reported for model comparison. NO 3 REACTIVITY MODEL obtained on a selected training set of 47 chemicals O 3 REACTIVITY MODEL obtained on a selected training set of 63 chemicals ATMOSPHERIC PERSISTENCE INDEX (ATPIN) The behavior of 66 heterogenous chemicals for which experimental k OH, k NO3 and k O3 values were available was studied in the Principal Component space. The first component was found to be the most important with 88.3% of explained variance; it’s easily interpreted as an indicator of the global tropospheric degradability of chemicals. The PC1 score can be proposed to rank organic chemicals according to the global tropospheric degradability (ATPIN) and this index is also modelled by theoretical molecular descriptors. 66 objects 4 variables R 2 =93.76% Q 2 LOO =92.53 Q 2 LMO(50%) =91.44% RMS=0.163 CONCLUSIONS  New predictive models for k OH k NO3 and k O3 are proposed.  The proposed models for the tropospheric degradation of chemicals take advantage of calculated imput parametres very simply and rapidly.  Genetic Algorithm is applied for Variable Subset Selection.  The models have good predictive powers verified by internal and external validations.  Reaction rate costants with OH, NO 3 radicals and ozone for new chemicals (even not yet syntetized) can be predicted.  The ranking of organic chemicals for tropospheric degradability was satisfactory, as realised by Principal Component Analysis on oxidation rate constants.  The first component highlights the atmospheric molecule behaviour; the PC1 score is modelled by theoretical molecular descriptors. This model can be used for the screening and ranking of chemicals according to their atmospheric persistence just starting from their structure. REFERENCES (1) Atkinson,R., 1989. J.Phys.Chem.Ref.Data. Monograph 1, 1-246; (2) Atkinson,R., 1991. J.Phys.Chem.Ref.Data., 20, 461-506; (3) Atkinson,R., 1984. Chem.Rev.,84, 437-470; (4) Todeschini R., Consonni V. and Pavan M., 2001. DRAGON – Software for the calculation of molecular descriptors, rel. 1.12 for Windows. Free download available at http://www.disat.unimib/chm; (5) CHEM 3D –Cambridge Soft, 1997, MA, USA; (6) Todeschini, R., 2001. Moby Digs - Software for multilinear regression analysis and variable subset selection by Geneitc Algorithm, rel. 2.3 for Windows, Talete srl, Milan (Italy); -Log k OH -Log k NO3 -Log k O3 Atmospheric Persistence DESCRIPTORS (in order of significance) Eiv : ionization energy (nucleophilicity) MATS1m : 2D-autocorrelation of Moran (dimension and shape) nDB : number of double bonds – nO : number of oxygen atoms H-048 : number of hydrogen atoms on carbon DESCRIPTORS (in order of significance) Eiv : ionization energy (nucleophilicity) ATS1m : 2D-autocorrelation of Moreau-Broto (shape and dimension) RTe : GETAWAY descriptor weighted by electronegativity (charge distribution) DESCRIPTORS (in order of significance) (HOMO-LUMO)GAP: reactivity - nAB : number of aromatic bonds AMW : average molecular weight - nDB : number of double bonds MATS7e and R3e : 2D- (Moran) and 3D- (GETAWAY) autocorrelations weighted by electronegativity (charge distribution) (7) Todeschini, R.; Mauri, A., 2000; DOLPHIN- Software for Optimal Distance-based Experimental Design rel 1.1 for Windows, Talete srl, Milan (Italy); (8) SCAN- Software for Chemometric Analysis, rel. 1.1 for Windows, Jerll. Inc., Standard, CA, 1992 ; (9) Leardi, R.; Boggia, R.; Terrile, M.,. J. Chemom., 1992, 6, 267-281; (10) Wold, S. Eriksson, L. Chemometric Methods in Molecular Design, 1995, VCH, Germany, 309-318; (11) Golbraikh, A. Tropsha, A., J. Mol. Graph and Mod., 2002, 20, 269-276; DESCRIPTORS (in order of significance) HOMO : highest occupied molecular orbital (nucleophilicity) nBnz : number of aromatic rings AMW : average molecular weight DELS : molecular electropological variation (charge distribution) OH REACTIVITY MODEL obtained on a selected training set of 306 chemicals