Publication Details |
Category | Text Publication |
Reference Category | Journals |
Title (Primary) | Assessing the reliability of QSAR predictions for structurally complex chemicals |
Author | Brown, T.N.; Stenzel, A.; Goss, K.-U. |
Source Titel | Abstracts of Papers of the American Chemical Society |
Year | 2013 |
Department | AUC |
Volume | 246 |
Page From | 64-ENVR |
Language | englisch |
UFZ wide themes | RU3; |
Abstract | Predicting environmentally relevant properties for chemicals with a complex structure can be a difficult task, both theoretically and empirically. One aspect that is often overlooked in QSAR development is that the pool of measured values available for training and validating a predictive model are commonly only those chemicals which are experimentally accessible or are of special interest. Due to the technical challenges of measuring various environmentally relevant properties of chemicals with complex structures, datasets are often biased towards simple chemical structures. However, there is a counter example to this in the pharmaceutical industry, where many of the chemicals of interest may be structurally quite complex. Both of these cases present challenges in the derivation and application of predictive relationships. In the first case, the properties of complex chemicals may not be readily inferred from the properties of simple chemicals, a problem further compounded by the sparsity of data for complex chemicals with which to validate the predictive models. In the second case the development of models is more difficult, because the basic structural elements that determine a chemical property may be obscured by too many complex chemical structures in the training dataset. Assessing the effects that these experimental biases have on the predictive power of developed QSARs requires the proper characterization of an applicability domain. Some basic partitioning systems, such as hexadecane-air partitioning, are used as examples to explore the effect of a biased dataset in QSAR development, and how the applicability domain assists in the proper interpretation of predicted values. These basic systems are relevant because they are in turn used as descriptors to characterize basic intermolecular interactions in empirical LSER equations, and recent experimental efforts have expanded the measured values to include both simple and quite complex chemicals. |
Persistent UFZ Identifier | https://www.ufz.de/index.php?en=20939&ufzPublicationIdentifier=14497 |
Brown, T.N., Stenzel, A., Goss, K.-U. (2013): Assessing the reliability of QSAR predictions for structurally complex chemicals Abstr. Paper Am. Chem. Soc. 246 , 64-ENVR |