Statistical techniques for screening experimental or literature chemical databases for compounds exhibiting potential environmental activity are becoming increasingly utilized in environmental analysis as pragmatic and economical complementary tools to enhance or augment costly traditional analytical procedures. Utilizing the predictive modeling approach, it is often argued, implicitly permits an unlimited number of chemicals to be screened for specific behavioral or physicochemical characteristics in a variety of environmental and biological matrices, consequentially conserving the financial resources for exhaustive testing, yet providing a methodology that helps to insure that questionable compounds are more thoroughly tested. Moreover, such techniques provide a database of exhaustive test results from which investigators and regulators can extract relevant information for further research or decision-making.

To assess the efficiency of statistical modeling methods for predicting chemical processes in the environment, a one-year exploratory study utilizing Quantitative Structure-Activity Relationship (QSAR) methodology to obtain linear model equations for estimating the rates of chemical hydrolysis of several organophosphorus (OP) pesticides in natural river waters has been conducted. This modeling effort specifically considers the effects of chemical structure on reactivity and utilizes connectivity parameters from graph theory as quantitative structural descriptors. Derived model equations were examined to establish whether quantitative correlations between fundamental molecular characteristics and observed hydrolytic properties were possible. Inconclusive results for a training set of six OP pesticides indicate that there are inherent weaknesses in molecular connectivity theory when applied to complex reaction parameters that require further exploration. The inherent complexity of most chemical reaction mechanisms and the indistinct influence of both adjoining and distant atoms in the molecular environment makes it difficult for a single descriptor, even one as widely successful as connectivity indices, to adequately account for definitive structural characteristics of molecules. It is apparent from results of this study that molecular connectivity indices alone are often not discriminating enough descriptors for procuring comprehensive structure-property relationships beyond a rather restricted range of structural variation, at least when characterizing chemical reaction parameters.