The second NIEHS Predictive-Toxicology Evaluation Project involves 30 NTP chemobioassays for carcinogenesis. Thus far it has generated 18 sets of predictions from 13 groups in 4 countries; 14 manuscripts were published together in an EHP Supplement. MODELS: Human- expert heuristic and the following six intelligent- computer-system models are under development: decision tree by induction, rule set from decision trees, back-propagation neural network, rule set from trained neural net, Bayesian-belief net, and inductive-logic programming. Each uses a fundamentally different approach to perform pattern- recognition analysis of learning sets, to identify specific biological & chemical features & relationships that may augment human- hypothesis formation about mechanistic pathways of chemotoxicity. The multiple-model/common training-set approach creates an ideal opportunity to evaluate model differences & by consensus analysis, identify & combine unique aspects of many models to provide one that predicts with greater confidence & perhaps greater accuracy. DATABASE COMPILATION and REPRESENTATION: This is an ongoing activity, because opportunities and success of the database-mining research approach are limited only by the availability of enough data of suitable quality. We compiled values on the following chemical attributes: Ashby structural alert, structural de-alert, SMILES code, 2-D structure, molecular weight, ClogP, highest-occupied and lowest-unoccupied molecular-orbital energies (HOMO & LUMO), and COMPACT ratio. MTD doses were converted to molar units. Representations that incorporate specific morphology@site information into our models, rather than just presence or absence of any lesion at each site, were developed.
