Statistical equivalence of prediction models of the soil sorption coefficient obtained using different log P algorithms.

The soil sorption coefficient normalized to the organic carbon content (Koc) is a physicochemical parameter used in environmental risk assessments and in determining the final fate of chemicals released into the environment. Several models for predicting this parameter have been proposed based on the relationship between log Koc and log P. The difficulty and cost of obtaining experimental log P values led to the development of algorithms to calculate these values, some of which are free to use. However, quantitative structure-property relationship (QSPR) studies did not detail how or why a particular algorithm was chosen. In this study, we evaluated several free algorithms for calculating log P in the modeling of log Koc, using a broad and diverse set of compounds (n = 639) that included several chemical classes. In addition, we propose the adoption of a simple test to verify if there is statistical equivalence between models obtained using different data sets. Our results showed that the ALOGPs, KOWWIN and XLOGP3 algorithms generated the best models for modeling Koc, and these models are statistically equivalent. This finding shows that it is possible to use the different algorithms without compromising statistical quality and predictive capacity.

An alternative approach for the use of water solubility of nonionic pesticides in the modeling of the soil sorption coefficients.

The collection of data to study the damage caused by pesticides to the environment and its ecosystems is slowly acquired and costly. Large incentives have been established to encourage research projects aimed at building mathematical models for predicting physical, chemical or biological properties of environmental interest. The organic carbon normalized soil sorption coefficient (K(oc)) is an important physicochemical property used in environmental risk assessments for compounds released into the environment. Many models for predicting logK(oc) that have used the parameters logP or logS as descriptors have been published in recent decades. The strong correlation between these properties (logP and logS) prevents them from being used together in multiple linear regressions. Because the sorption of a chemical compound in soil depends on both its water solubility and its water/organic matter partitioning, we assume that models capable of combining these two properties can generate more realistic results. Therefore, the objective of this study was to propose an alternative approach for modeling logK(oc), using a simple descriptor of solubility, here designated as the logarithm of solubility corrected by octanol/water partitioning (logS(P)). Thus, different models were built with this descriptor and with the conventional descriptors logP and logS, alone or associated with other explanatory variables representing easy-to-interpret physicochemical properties. The obtained models were validated according to current recommendations in the literature, and they were compared with other previously published models. The results showed that the use of logS(p) instead of conventional descriptors led to simple models with greater statistical quality and predictive power than other more complex models found in the literature. Therefore, logS(P) can be a good alternative to consider for the modeling of logK(oc) and other properties that relate to both solubility and water/organic matter partitioning.

The effect of different log P algorithms on the modeling of the soil sorption coefficient of nonionic pesticides.

Collecting data on the effects of pesticides on the environment is a slow and costly process. Therefore, significant efforts have been focused on the development of models that predict physical, chemical or biological properties of environmental interest. The soil sorption coefficient normalized to the organic carbon content (Koc) is a key parameter that is used in environmental risk assessments. Thus, several log Koc prediction models that use the hydrophobic parameter log P as a descriptor have been reported in the literature. Often, algorithms are used to calculate the value of log P due to the lack of experimental values for this property. Despite the availability of various algorithms, previous studies fail to describe the procedure used to select the appropriate algorithm. In this study, models that correlate log Koc with log P were developed for a heterogeneous group of nonionic pesticides using different freeware algorithms. The statistical qualities and predictive power of all of the models were evaluated. Thus, this study was conducted to assess the effect of the log P algorithm choice on log Koc modeling. The results clearly demonstrate that the lack of a selection criterion may result in inappropriate prediction models. Seven algorithms were tested, of which only two (ALOGPS and KOWWIN) produced good results. A sensible choice may result in simple models with statistical qualities and predictive power values that are comparable to those of more complex models. Therefore, the selection of the appropriate log P algorithm for modeling log Koc cannot be arbitrary but must be based on the chemical structure of compounds and the characteristics of the available algorithms.