Computerassisted Sar Predictive Models And Systems

The traditional mechanism-based SAR prediction of potential carcinogens attempts to consider all current knowledge of the chemical and biological processes in tumorigenesis. Yet, there are considerable knowledge gaps in understanding the complexity of the interaction between every chemical structure within a biological system. Predictions of carcinogenic or mutagenic activity often have to be based on limited knowledge and uncertainty of effects of uncommon structural features on the metabolism, toxicokinetics, and postulated mechanisms of the chemicals. Furthermore, this approach, which relies heavily on individuals with cumulative knowledge of and experience in the chemical and biological mechanisms in carcinogenesis, is difficult for nonexperts to put into routine practice. Recent advances in computer technology coupled with increasing economic and societal pressures to reduce animal testing have motivated researchers to develop comprehensive models for screening and predicting carcinogenic or mutagenic activity of the large number of untested chemicals that have diverse structures and, possibly, different modes of action. Carcinogenicity and mutagenicity databases contain a wealth of information encoded in their chemical structures and corresponding biological activities. Analysis of large databases by well-designed computer models has the potential of discerning subtle structural modifiers to biological activity and, thus, the potential for improving current predictive capability. Computer models, such as "expert systems" ("artificial intelligence" systems) can also codify and encapsulate human expertise, reduce or eliminate error and inconsistency, and help nonexperts arrive at expert judgement regarding carcinogenic potential of chemicals.

Using the concept of quantitative structure-activity relationships (QSAR) developed by Hansch (22, 23), a number of commercially available computerassisted predictive models for toxicologic assessment have been developed. In general, the development of computer-assisted QSAR models involves the graphical entry and storage of structures, generation of molecular structure descriptors, analysis of the descriptors, development of quantitative relationships between descriptors, and biological responses using multivariate statistical or pattern recognition methods. Structural descriptors used for model development include physicochemical, topologic, geometrical, electronic, and other quantum mechanical properties of the molecules.

Two of the most well-known computer-assisted models for carcinogenicity and mutagenicity prediction are the MultiCASE/CASE (Computer Automated Structure Evaluator) (24-26) and the TOPKAT (Toxicity Prediction by Komputer Assisted Technology) (27, 28). Both of these computerized predictive models are based on pattern recognition analysis of learning sets that are carcinogenicity databases of chemicals with noncongeneric structures. They differ from each other primarily in the way substructural fragments are identified and incorporated in the systems.

The CASE/MultiCASE model is completely automated in the generation and inclusion of its molecular substructural fragment descriptors from the learning database. In TOPKAT, the substructural fragments are chosen from a preselected library of more than 3,000 potentially significant fragments compiled on the basis of general organic chemistry functionality and mechanistic considerations. In addition to using substructural fragments as molecular descriptors, TOPKAT and MultiCASE also use molecular topologic indices, molecular shape indices, electronic properties, and partial atomic charges of chemical structures as modifiers from the learning databases for their development. When a new chemical is evaluated, the model will search its structure for the existence of a discriminating feature or a biophore. If a discriminating feature or a biophore is found, the model will then search for the presence of potential modulators to arrive at a projected value for its potency. The statistically based, correlative approach used by these computer-assisted predictive models has the advantage of minimizing human expert input and bias. The knowledge limit and predictive range, determined totally by the information contained within the learning databases, however, are still relatively small for statistical analyses and prediction of the large number of untested chemicals with diverse structures and modes of action. Furthermore, as these models were derived from existing carcinogenicity databases, uncertainties and inaccuracies associated with rodent carcinogenicity studies and with interpretation and assignment of chemicals of carcinogenicity "calls" inherent in such databases are all reflected in the SAR prediction models. A good illustration of the former can be provided by the report of poor performance of the standard MultiCASE and other programs to predict the carcinogenicity potential of pharmaceuticals due to poor coverage for drug molecules and inadequate representation of the molecular diversity of drugs; to correct this deficiency, a new, enhanced FDA-OTR/MCASE software program has been developed with the inclusion of more than 1,000 compounds from the FDA-OTR Carcinogenicity Database to the MultiCASE learning database (29).

Correlative SAR models also have the deficiency of rationalizing molecular mechanisms related to carcinogenicity. Hence, a number of other computerassisted SAR predictive models also incorporate some mechanistic considerations and expert judgement, as well as using statistical association, in their development. Examples include the ke carcinogenicity model of Benigni (30, 31) and the COMPACT (Computer-Optimized Molecular Parametric Analysis of Chemical Toxicity) model of Parke (32, 33). The ke carcinogenicity model measures the ke (electrophilicity) of a chemical, and the prediction is based on the relevance of the ke to carcinogenicity. The COMPACT model has been developed on the basis of specific physicochemical conditions (e.g., degree of planarity of the molecule) for the metabolic activity among a particular subfamily of cytochrome P-450 oxidases as a predictor of potential carcinogenic activity. It is a molecular orbital method that predicts whether a chemical may be metabolized by a specific cytochrome P-450 oxidase, that is, CYP1, CYP2E, or CYP3, to form a reactive intermediate that leads to carcinogenicity. One major weakness of these models is that they are tied to a single mechanistic hypothesis and are not modeled for discovering new mechanistic insight that may lead to greater predictive capability.

Rather than using a statistically based, correlative approach, the OncoLogic Cancer Expert System has been developed based on knowledge rules that represent the formalized, codified, and organized SAR knowledge of human experts (34-36). The input of this rule-based computerized expert system includes the chemical structure as well as all available chemical, biological, and mechanistic information (e.g., physicochemical properties, chemical stability, route of exposure, bioactivation and detoxicification, genotoxicity, and other supportive data) critical to the evaluation of carcinogenic potential. In contrast to other SAR predictive models that are not based on mechanistic considerations or use only single mechanistic assumption, the OncoLogic Cancer Expert System uses different sets of knowledge rules specific for different chemical classes or subclasses to account for their different modes of action. This approach maximizes the accuracy of predicting not only organic compounds of diverse structures or substructures, but also other types of chemical substances such as fibers, metals, and polymers. The DEREK (Deductive Estimation of Risk from Existing Knowledge) is another knowledge-based expert system developed for qualitative prediction of carcinogenicity of chemicals. The DEREK system makes its predictions based primarily on a small set of rules, each of which describes the relationship of a structural feature ("toxicophore") to carcinogenicity (37, 38). Because knowledge rules are based on the expertise and intuition of a few individuals, rule-based expert systems are more biased in their perception of SAR knowledge than statistically based, correlative models. Like the traditional mechanism-based SAR approach, current knowledge gaps in mechanism and mode of action also limit the range of applicability of these computerized expert systems. The carcinogenicity of only chemical congeners within chemical classes rich in mechanistic data can be predicted with a high level of confidence.

There are a number of attractive features for computer-assisted SAR models, but difficulties and problems exist involving model development, validation, confirmation, and acceptance issues (39). Nonetheless, computer-assisted predictive models can function as valuable tools for chemical screening and hazard identification if they are applied with adequate oversight and scrutiny. As part of the Predictive-Toxicology Evaluation Project of the U.S. National Institute of Environmental Health Sciences (NIEHS), the state of the art of predicting rodent carcinogenesis and the strengths and weaknesses of different predictive methods have been evaluated (40, 41). The results and conclusions of the first round of prospective predictions made by several different methods on a set of 44 previously untested, noncongeneric chemical substances are: (a) despite the different approaches, there was good agreement among the predictions made for some chemicals that produced unambiguous bioassay results; (b) SAR models that are based primarily on chemical structure and that ignore biological attributes did not perform as accurately as models that use biological information; and (c) models such as human experts that use more extensive and varied information performed better than models based on one or two attributes to represent chemical carcinogenesis.

The strengths and weaknesses of various SAR approaches for carcinoge-nicity prediction by computer-assisted models have also been reviewed (42).

10 Ways To Fight Off Cancer

10 Ways To Fight Off Cancer

Learning About 10 Ways Fight Off Cancer Can Have Amazing Benefits For Your Life The Best Tips On How To Keep This Killer At Bay Discovering that you or a loved one has cancer can be utterly terrifying. All the same, once you comprehend the causes of cancer and learn how to reverse those causes, you or your loved one may have more than a fighting chance of beating out cancer.

Get My Free Ebook

Post a comment