Prediction of Henry's Law Constants by a Quantitative Structure Property Relationship and Neural Networks

Abstract

Multiple linear regression analysis and neural networks were employed to develop predictive models for Henry's law constants (HLCs) for organic compounds of environmental concern in pure water at 25 degrees C, using a set of quantitative structure property relationship (QSPR)-based descriptors to encode various molecular structural features. Two estimation models were developed from a set of 303 compounds using 10 and 12 descriptors, one of these models using two descriptors to account for hydrogen-bonding characteristics explicitly; these were validated subsequently on an external set of 54 compounds. For each model, a linear regression and neural network version was prepared. The standard errors of the linear regression models for the training data set were 0.262 and 0.488 log(H(cc)) units, while those of the neural network analogues were lower at 0.202 and 0.224, respectively; the linear regression models explained 98.3% and 94.3% of the variance in the development data, respectively, the neural network models giving similar quality results of 99% and 98.3%, respectively. The various descriptors used describe connectivity, charge distribution, charged surface area, hydrogen-bonding characteristics, and group influences on HLC values.

A Fuzzy ARTMAP‐Based Quantitative Structure—Property Relationship (QSPR) for the Henry′s Law Constant of Organic Compounds.

“…If a given chemical in a test does not have a matching category, developed in the training phase, the fuzzy ARTMAP will match the compound with the closest available recognition category according to a tolerance set by the vigilance parameter. For example, during model evaluation with the test set, octanal (logH ) -1.68), an aldehyde with a molecular formula of C 8 H 16 O, was classified with 2-octanone (logH ) -2.11), a ketone also with a molecular formula of C 8 H 16 O. The above examples suggest that further improvements to the fuzzy ARTMAP logH QSPR would require a data set that contains a larger number of compounds per class and possibly a refined set of molecular descriptors to allow a greater ability to differentiate among complex or apparently very similar structures.…”

“…It is interesting to note that Brennan et al 6 suggested that predictions of Henry's Law constant within a factor of 2.5 are a reasonable for many environmental applications given the variability among measured data, where standard deviations can range from less than 0.05 to about 0.5 logH units. 16 The performance of the present logH QSPR models can also be assessed based on error analysis for specific chemical groups (Table 4). It is noted that the errors for the majority of the different chemical groups are within the same order of magnitude, for each respective model; however, errors for the fuzzy ARTMAP are generally about 2 orders of magnitude lower than for the back-propagation-based QSPR.…”

“…In contrast, the present back-propagation and fuzzy ARTMAP based models performed with average absolute errors of 0.13 and 0.27 logH units, respectively for a Henry's law constant range of -6.72 e logH e 2.87. More specific comparisons of the performance of the present to previously published logH QSPRs, 4,5,[15][16] for compounds common with the present data set, are provided in Table 5 and Figure 5. The predictions of the back-propagation and fuzzy ARTMAP models, for 277 compounds common with the data set of Nirmalakhandan and Speece, 4,5 were within an average absolute and maximum errors and standard deviation of 0.26 (19.5%), 1.3 (127%) and 0.25 (21.4%) and 0.03 (3.6%), 0.5 (131%), and 0.07 (13.2%) logH units, respectively.…”

“…Unfortunately, group interaction parameters for UNI-FAC and reliable vapor pressure data are often lacking for chemicals of environmental interest, as well as for many heterocyclic aromatic compounds and compounds with functional groups containing sulfur and phosphorus. [7][8][9][10] Group contribution 11,12 and quantitative structure-property relations (QSPRs) [4][5][6][13][14][15][16] have been popular estimation methods of the Henry's Law constant for organic compounds. For example, Hine and Mookerjee 11 proposed bond contribution and group contribution schemes for logH, based on data sets of 263 and 212 compounds (-5 <logH <2), with a reported standard deviations of 0.42 and 0.12 logH units, respectively, for the above two sets.…”

“…The advantage of NNs, over classical regression analysis methods, is their inherent ability to incorporate nonlinear relationships between chemical structural parameters and physicochemical properties. [16][17][18][19][20][21] Neural network/QSPR models for estimating the Henry's Law constant for a data set of 357 organic compounds (-7.08 e logH e 2.32) have been recently reported by English and Carroll (2001). 16 The above authors reported QSPRs based on 12-4-1 and 10-3-1 backpropagation neural network architectures (trained using 303 compounds) that performed with absolute errors of 0.237 and 0.281 logH units for the test set (54 compounds), respectively.…”