Expert system chemical structure prediction

Expert systems have been defined as any formal systems , which make predictions about the toxicity of chemicals. All expert systems for the prediction of toxicity are built on experimental data and/or rules derived from such data (Dearden 2003). The expert systems can be further divided into two subclasses based on the method of generating rules. The one method is a knowledge- or rule-based expert system, for which experts/toxicologists create rules based on a list of structural features that have been related to a specified toxicity (Durham and Pearl 2001). An example of a typical knowledge- or rule-based system is DEREK, which will be described later. 

J. P. Hickey, A. Aldridge, D. R. May Passino, A. M. Frank, Expert System Predicts Aquatic Toxicityfrom Contaminant Chemical Structure,NMoa-A Fisheries Research Center-Great Lakes, U.S. Fish and Wildlife Service, Ann Arbor, Mich., 1991 Ibid., Drug Information Journal 26, 487 (1992). 

As computing capability has improved, the need for automated methods of determining connectivity indexes, as well as group compositions and other structural parameters, for existing databases of chemical species has increased in importance. New naming techniques, such as SMILES, have been proposed which can be easily translated to these indexes and parameters by computer algorithms. Discussions of the more recent work in this area are available (281,282). SMILES has been used to input Contaminant structures into an expert system for aquatic toxicity prediction by generating LSER... 

Chemical metabolism can be described qualitatively or quantitatively. Many scientists can make qualitative predictions of the likely excretion products or blood plasma metabolites in mammals, or a particular animal including man, based on accumulated knowledge and experience. Such knowledge, in its raw form, generally consists of structure-metabolism relationships that are frequently expressible as qualitative structure-based rules that may be encoded into computer-based expert systems (see Chapter 9 for a full definition). Examples of such systems, in their more fully developed commercial forms, are discussed toward the end of this chapter. 

Figure 13.11 Overview diagram of the NCTR Four-Phase approach for priority setting. In Phase I, chemicals with molecular weight < 94 or > 1000 or containing no ring structure will be rejected. In Phase II, three approaches (structural alerts, pharmacophores, and classification methods) that include a total of 11 models are used to make a qualitative activity prediction. In Phase III, a 3D QSAR/CoMFA model is used to make a more accurate quantitative activity prediction. In Phase IV, an expert system is expected to make a decision on priority setting based on a set of rules. Different phases are hierarchical different methods within each phase are complementary.

The two major independent in silico methods for the prediction of toxicity are quantitative-structure-activity-relationship (QSAR) and expert systems (e.g. DEREK, MultiCASE). QSAR means the quantitative relationship between a chemical structure and its biological/ toxicological activity with the help of chemical descriptors that are generated from the... 

Another approach to predict toxicity is basing on structure-activity-relationship (SAR), which means the qualitative relationship between a specific chemical structure and their biological/toxicological activity, e.g. the expert system DEREK is based on SAR prediction. In SAR the occurrence of specific substructures in a molecule are correlated to be responsible and necessary for a biological/toxicological activity. 

The relationship between chemical structure and activity has been recognized from early on, as the examples in Sec. 10.3.2 illustrate. Tbday, computerized expert systems allow the virtual screening of millions of possible structures with the objective to find the best candidates for development. However, the accuracy of such in-silico predictions is still low and it will take time before they can replace experimental discovery methods. 

In silico tools make a significant contribution to the SAR-based early identification of potential toxicity. An increasing volume of published preclinical and clinical toxicity data are collected and used to build structure-related searchable databases. These expert knowledge databases can analyze chemical structures and match them with potential mechanisms of toxicity. DEREK for Windows (Lhasa Ltd.)39 is one of such broadly used knowledge-based expert systems to provide toxicology alerts for new compounds. Although certainly not comprehensive, numerous efforts have been made to predict hepatotoxicity. Recently,... 

In silico models (or expert systems ) have also been developed. These are computer software-based structure-activity relationship and quantitative structure-activity relationship analyses of data libraries of acute toxicity data developed for use in evaluating and predicting the acute oral and inhalation toxicity potential of a chemical or drug. 

Cronin MTD. The use by governmental regulatory agencies of quantitative structure-activity relationships and expert systems to predict toxicity. In Cronin MTD, Livingstone DJ, editors, Predicting chemical toxicity and fate. Boca Raton, FL CRC Press, 2004. p. 413-27. 

The software now uses structurally intrinsic parameters for only one QSAR model (LSER) and the results are used to predict one property (acute toxicity) to four aquatic species by one mechanism (nonreactive, non-polar narcosis) however, we intend to continue to refine our equations as databases grow, incorporate other models, predict other properties, and include other organisms. We will attempt to differentiate between modes of toxic action and improve our estimates accordingly. For the widely divergent classes of chemicals and types of environmental behavior, no one model will best describe every situation and no one species is the optimal organism to monitor. As the software evolves, the expert system should choose the best model based on the contaminant, the species, and the property to be predicted (e.g., toxicity or bioaccumulation). In addition, we envision an interactive screen system for data entry that will bypass the SMILES notation and allow the user to describe the molecule by posing a series of questions about the compound s backbone and functional groups. The responses will translate directly into values of LSER variables. 

---