P045 (DIM-1117-793826) Time Dependent Biodegradation Model.
Start time: 8:00 AM
Mekenyan, Ovanes¹, Dimitrov, Sabcho¹, Nedelcheva, Daniela¹, Reuschenbach, P², Silvani, M², ¹ Laboratory of Mathematical Chemistry, University “Prof. As. Zlatarov”, Bourgas, Bulgaria² BASF Aktiengesellschaft, Product Safety – Regulations, Toxicology and Ecology, Aktiengesellschaft, Germany
The rate of biodegradation of organic chemicals necessary for regulatory purposes as well as for industry requires development of a model for predicting the extent of biodegradation at different time frames, including half-live and the 10 days window. Conceptually this implies expressing the rate of catabolic transformations as a function of time. The attempt to explain time dependence of biodegradation will be presented. The biodegradation model was formulated assuming the first order kinetics for catabolic transformations. The mathematical formalism combined with simulator of metabolic logic of catabolism allows simulation of integral biodegradation data (BOD, CO2 production) as well as metabolism (biodegradation pathways, amount of metabolites, etc.) at different time frames. The performance of the simulator of the metabolic logic of catabolism based on kinetic biodegradation data provided by industry will be demonstrated.

P046 (DIM-1117-799315) Acute Aquatic Toxicity and Species Extrapolation Models.
Start time: 8:00 AM
Mekenyan, Ovanes¹, Dimitrov, Sabcho¹, Koleva, Yana¹, Schultz, T², ¹ Laboratory of Mathematical Chemistry, Bourgas University “Assen Zlatarov”, Bourgas, Bulgaria² Department of Comparative Medicine, College of Veterinary Medicine, The University of Tennessee, P. O. Box 1071, Knoxville, Ten, USA
This investigation examines the quantitative structure-activity relationships for aquatic toxic potency of variety chemicals with different mechanisms of toxic action, such as non-polar narcotics, esters and amine narcotics, phenols, anilines, aldehides, etc. The models proposed here are based on the comparative study of ciliate (Tetrahymena pyriformis) population growth impairment data, fish (Pimephales promelas, Poecilia reticulate, Carassius auratus) mortality data, crustacean (Daphnia magna) efficient (48h immobilisation) and mortality data, tadpole (Rana temporaria) minimum narcosis concentration, mosquito larvae (Culex tarsalis) mortality data, as well as data for other aquatic animals. The modeling method for explaining toxicity of non-polar narcotcs is based on the response-surface concept, using bioconcentration (log BCF) and orbital electrophilicity (E_LUMO) as global molecular descriptors for accumulation and reactivity of chemicals, respectively. For covalently acting chemicals the global electrophilicity is replaced by local molecular descriptors assessing reactivity of toxicophores. The statistical quality of the derived interspecies models was found to be comparable with those of QSARs fitted on separate endpoints. The interspecies models are easy to interpret and provide a method for the extrapolation of toxic potency from one organism to another.

P047 (RED-1117-827178) Application of the Narcosis Target Lipid Model to Wastewater Treatment Plant Microogranisms.
Start time: 8:00 AM
Redman, A¹, McGrath, J¹, Parkerton, T², Di Toro, D³, ¹ HydroQual, Inc.² ExxonMobil³ University of Delaware
The narcosis target lipid model (NTLM) was developed to predict the toxicity of chemicals that act via narcosis to aquatic organisms. The model accounts for variations in toxicity due to species sensitivities and chemical differences. The model is based on the hypothesis that lipid is the site of action within the organism. Previous work has focused on the application of this model to fish, algae and invertebrates. The objective of this study was to determine if this model could logically be extended to predict toxicity endpoints for microorganisms common to wastewater treatment plants (WWTPs). Toxicity datasets were gathered for twelve different species, including MICROTOX, ciliated protozoa, nitrifying bacteria and several microbial consortia derived from WWTP activated sludge. Linear relationships between Log(EC50) and Log(K_ow) were derived for fourteen individual datasets. For twelve datasets, the slopes of the individual log(EC50)-log(K_ow) relationships were found to be not statistically different from the universal narcosis slope derived from earlier analysis of toxicity data for aquatic organisms. The universal narcosis slope was applied to the individual data sets and critical target lipid body burdens (CTLBB) were determined for each WWTP species. No difference in species sensitivity was found. Acute to Chronic Ratios (ACR) were derived from available data and were found to overlap the previously reported range of ACRs. These data were used to calculate the Hazardous Concentration to 5% of the tested species, a statistical extrapolation comparable to a predicted-no-effect concentration (PNEC).

P048 (BIR-1117-819074) A New Method for Estimating Beef and Milk Biotransfer Factors.
Start time: 8:00 AM
Birak, P¹, Yurk, J², ¹ RTI International, RTP, NC, USA² USEPA Region 6, Dallas, TX, USA
In multi-media risk assessments, ingestion of contaminated animal products is a commonly considered pathway. Such assessments, typically predict chemical transfer into beef and milk using a biotransfer factor (BTF). BTFs relate chemical intake rates to concentrations at the point of exposure. Unfortunately, experimental data for deriving beef and milk BTFs is limited to a handful of chemicals. For most compounds, risk assessors estimate BTFs based on chemical properties. Existing predictive equations, based on linear regressions, over predict BTFs for high log Kow compounds (i.e., >6). The goal of this research was to develop an improved methodology for predicting chemical BTFs for beef and milk. The primary objectives were to (1) ensure that the methodology was transparent and easy to implement, (2) use only chemical properties for which data were readily available , and (3) demonstrate improvements in the accuracy of biotransfer predictions over current methods. Additionally, we developed an extensive database of biotransfer data. Though alternate chemical properties were considered, BTFs are predicted using log Kow. Chemicals capable of undergoing hydrolysis or oxidation reactions were excluded from regression analyses. Remaining data were fit using a polynomial regression to improve BTF predictions for high log Kow compounds.

P049 (KOL-1117-830442) Increased endocrine activity of xenobiotic chemicals as mediated by metabolic activation.
Start time: 8:00 AM
Kolanczyk, R¹, Tapper, M¹, Nelson, B², Wehinger, V², Denny, J¹, Kuehl, D¹, Sheedy, B¹, Mazur, C³, Jones, J³, Schmieder, P¹, ¹ US EPA, ORD/NHEERL, Mid-Continent Ecology Division, Duluth, Minnesota, USA² US EPA Student Services Contractors, Duluth, Minnesota, USA³ US EPA, ORD, NERL, Ecosystems Research Division, Athens, Georgia, USA
EPA is faced with long lists of chemicals that need to be assessed for hazard. This research is part of a larger effort to develop in vitro assays and QSARs applicable to untested chemicals on EPA inventories through study of estrogen receptor (ER) binding and estrogen mediated gene expression in fish. The current effort investigates metabolic activation of chemicals resulting in increased estrogenicity. Phenophthalin (PLIN) was shown not to bind trout ER in a competitive binding assay but vitellogenin expression was induced in trout liver slices exposed to 10^-4 and 10^-3.7 M PLIN. Phenolphthalein (PLEIN), a metabolite of PLIN, was subsequently determined to be formed when slices were exposed to PLIN. PLEIN binds rtER with a relative binding affinity (RBA) of 0.020%. Slices exposed to PLEIN expressed vitellogenin mRNA at 10^-4.3 10^-4 and 10^-3.7 M, with no detectable PLIN present. Thus, vitellogenin expression noted in PLIN slice exposures was explained by metabolism to PLEIN in trout liver slices. A second model chemical, 4,4′-diaminodiphenylmethane (MDA) was not shown to bind rtER, but did induce vitellogenin mRNA production in tissue slices at 10^-4.3 10^-4 and 10^-3.7 M in amounts nearly equal to reference estradiol induction, thus indicating metabolic activation of MDA. A series of experiments were performed to identify a potential metabolite responsible for the observed increase in activity. Hydroxylamine-MDA, nitroso-MDA, azo-MDA, or azoxy-MDA were not observed. Acetylated-MDA was observed and tested in both ER-binding and tissue slice vitellogenin induction assays. Comparisons of Phase I metabolic activation in trout and rat liver microsomes is also presented to elucidate cross species similarities and to enhance predictive models. This abstract does not necessarily reflect U.S. EPA policy.

P050 (TAP-1117-829199) In vitro trout assays for interpreting potential estrogenicity of industrial chemicals.
Start time: 8:00 AM
Tapper, M¹, Denny, J¹, Kolanczyk, R¹, Sheedy, B¹, Nelson, B², Wehinger, V², Schmieder, P¹, ¹ US EPA, NHEERL, Mid-Continent Ecology Division, Duluth, MN, USA² US EPA Student Services Contractor, Duluth, MN, USA
Two in vitro rainbow trout assays, cytosol estrogen receptor competitive binding, and vitellogenin (VTG) liver slice mRNA expression were used to interpret potential estrogenicity of industrial chemicals. This work is part of a larger project to develop in vitro approaches and QSAR models applicable to untested chemicals on EPA inventories. Over 150 chemicals have been tested in the competitive binding assay, with the focus on interpreting the relevance of low relative binding affinities (RBA) found for the majority of industrial chemicals of EPA concern; the RBA range of interest is from 0.5 to 0.000026%. Additionally, >50 chemicals were tested in the male trout liver VTG mRNA expression assay, each selected to enhance interpretation of binding curves. Chemicals chosen for confirmation in slice VTG tests included many that yielded characteristic competitive binding curves (shape analogous to E₂), non-binders (no indication of displacement) and many chemicals yielding non-characteristic curves (e.g., steep, low efficacy). Several low RBA chemicals tested to assess relevance of low RBA to ability to induce vitellogenesis are presented. The liver slice model proved to be an effective tool for determining relevance of low binding affinity, chemicals with RBA's as low as 0.0004% induced VTG expression at 50 to 100% of the maximum response to estradiol, as well as clarifying the relationship between chemical concentration and toxicity. Furthermore, the slice model was valuable for interpretation of abnormal binding curves, identifying both estrogens and non-estrogens with steep curves, and determining that binding displacement > 50% was needed to elicit VTG expression. In two instances of a chemical exhibiting <50% binding efficacy but increased VTG expression, a metabolite was found responsible for the induction of gene expression. Therefore, in addition to binding affinity, factors such as chemical uptake, toxicity and metabolism ultimately determined estrogenic potency in trout liver slices.

P051 (TAN-1117-507801) Construction of Carcinogenicity Database CAESAR.
Start time: 8:00 AM
Tanabe, K¹, Ohmori, N¹, Ono, S¹, Suzuki, T², ¹ Chiba Institute of Technology, Narashino, Chiba, Japan² Toyo University, Bunkyoku, Tokyo, Japan
Among chemical substances available in market, quite few are ascertained on safety, and reliable data on chemical safety are few and sparse. So, it is necessary to construct a database which collect reliable data on chemical safety. Since animal test to get data on chemical safety takes much cost and period, it is impossible to get safety data on every unascertained chemical substance by animal test. Here a computer prediction with quantitative structure-activity relationship (QSAR) is meaningful for screening. Therefore many attempts based on QSAR models for estimating the chemical safety such as carcinogenicity have been made. Several systems for predicting carcinogenicity were applied to the Predictive Toxicology Challenge contest, but their performances were all low. This is because the chemical safety data, especially of carcinogenicity, consist of two discrete values (carcinogenic or non-carcinogenic) for diverse chemical substances. QSAR approaches succeeded in predicting carcinogenicity of congenial substances, using experimental data with relative carcinogenicity, but failed to predict carcinogenicity of non-congenial substances using two discrete carcinogenicity data. So, if we construct a chemical safety database, especially for carcinogenicity consisting of relative strength or reliability, QSAR approach may succeed in predicting the safety for non-congenial substances. We are constructing a chemical safety database named CAESAR (Computer-Aided Evaluation of Chemical Safety with QSAR), which consists of two databases. One contains reliable, critically reviewed, experimental hazard data with relative hazard strength or reliability on selected chemical substances. Another contains hazard data predicted with QSAR for numerous chemical substances commercially available. As a first step, we are constructing a carcinogenicity database. Reliable experimental carcinogenicity data on about 1,000 chemical substances are collected from various sources such as IARC, NTP and others, and their reliabilities are ranked into nine categories. Reliabilities of carcinogenicity for diverse chemical substances are predicted using those experimental data with a neural network method which we already developed.

P052 (TOD-1117-800366) A QSAR Evaluation of AR Binding Affinity of Chemicals.
Start time: 8:00 AM
Mekenyan, O¹, Todorov, M, Serafimova, R ¹, Schmieder, P², Aladjov, H², ¹ Laboratory of Mathematical Chemistry, Bourgas As. Zlatarov University, Bourgas, Bulgaria² U.S. Environmental Protection Agency, Mid-Continent Ecology Division, 6201 Congdon Blvd., Duluth, MN, USA
Some of the environmental and industrial chemicals can interact with the androgen receptor (AR) by mimicking the functions of natural hormones. The multiparameter formulation of COmmon REactivity PAttern (COREPA) approach was used to describe the structural requirements for eliciting androgen potency. Structurally diverse training data set containing 202 chemicals was obtained from the National Center for Toxicology Research (NCTR, US). In agreement with previous models, chemical affinities for the ratAR were related to distances between nucleophilic sites and structural features describing electronic interactions between the receptor and ligands. A single binding mechanism was identified similar to the steroidal A-B electronic mechanism identified for ER binding activity. The COREPA model predicting AR binders with relative binding affinity (RBA) higher then 10% (all steroids) contains: a distance screen of 10.55-10.90 A between hydroxyl O and sp2-hybridized O-atom combined with a classification node containing 2 discriminated parameters - maximal donor delocalizability and LogKow. Almost same decision tree was obtained for the AR activity range 0.1
P053 (HUG-1117-720112) Comparison of QSAR-based predictive software used for physical/chemical property estimation.
Start time: 8:00 AM
Hughes, D¹, MacPherson, H¹, Bose, R¹, ¹ Health Canada, New Substances Assessment and Control Bureau, Ottawa, Ontario, Canada
This preliminary study was aimed at comparing and evaluating QSAR software used for physical/chemical property predictions in the pre-market regulatory evaluation of new commercial chemical substances. In addition to test data, a variety of such estimates are utilized in the exposure and risk assessment of these substances. We wished to examine inter-estimate variability between the different software and additionally to compare the estimates with laboratory results. Estimated values for physical chemical endpoints generated by QSAR-based predictive software (ACD, EPISUITE, PALLAS, and SPARC which use a variety of QSAR methodologies) were utilized for new substances notified for assessment under the Canadian Environmental Protection Act. The sample consisted of several chemical categories of new substances (hindered amines, esters, acetamides, napthalenesulfonic acids, alcohols, amines, carboxylic acids, aldehydes, and urea derivatives) with varied uses (lubricants, catalysts, dyes, thickeners, corrosion inhibitors, coating additives, rheology additives, flame retardants, light stabilizers, carrier fluids etc.) for which ACD estimates were submitted by notifiers; we also generated estimates using other software, for comparison. The endpoints reported here include pKa (-5 to 18), log Kow (-4 to 27), and WS (10E-27 to 106 mg/L). The pKa estimates were found to demonstrate good correlation (SPARC vs ACD; R² = 0.93). The QSARs for log Kow were also found to yield reliable predictions (ACD vs SPARC; R² = 0.77; PALLAS vs KOWWIN; R² = 0.78, ACD vs KOWWIN; R² = 0.83, SPARC vs KOWWIN; R² = 0.88). The water solubility estimates were found to be the least consistent of the property estimates examined (SPARC vs WSKOW; R² = 0.65). The online software SPARC was not able to calculate properties for certain complex molecules and had usability problems. Any of these software examined would make comparable and acceptable predictions for most chemicals for the pKa and log Kow endpoints.

P054 (MEK-1117-808477) Existing Chemicals in Regulatory Agencies of North America and Europe: 3D-Databasing Accounting for Chemicals Flexibility and Quantum-chemical Evaluation.
Start time: 8:00 AM
Mekenyan, O¹, Nikolov, N², Grancharov, V¹, Stoyanova, G¹, Pavlov, T¹, ¹ Laboratory of Mathematical Chemistry, University “Prof. As. Zlatarov”, Bourgas, Bulgaria² CLBME - Bulgarian Academy of Sciences, Bl. 105 Acad. G. Bonchev, 1113, Bulgaria, Sofia
The present inventory of existing chemicals in regulatory agencies in North America and Europe, encompass chemicals of European Chemicals Bureau (EINECS, with 61428 discrete chemicals), Danish EPA (159445 chemicals), US EPA (TSCA, 56884 chemicals), US EPA (HPVC, 10271 chemicals) and Pesticides active/inactive ingredients of US EPA (1380 chemicals), OECD (HPVC, 4625 chemicals) Environment Canada (DSL, 10857 chemicals) and Japanese METI (14289) were combined in CD-EC. The total number of unique chemicals from all these databases exceeded 183 000. Defined and undefined chemical mixtures and polymers are handled, along with discrete chemicals. The database manager provides storage and retrieval of chemical structures with 2D and 3D data, accounting for molecular flexibility by using representative sets of conformers for each chemical. The electronic and geometric structures of all conformers are quantum-chemically optimized and evaluated. Hence, the database contains over 3.5 million 3D records with hundreds of millions of descriptor data items at the levels of structures, conformers, or atoms. The platform contains a highly developed search subsystem-search is possible on CAS numbers, names, 2D and 3D fragment search, structural, conformational, or atomic properties, affiliation in other chemical databases, structure similarity, logical combinations, saved queries, search result exports. Models (collections of logically related descriptors) are supported, including information on models author, date, bioassay, organ/tissue, conditions, administration etc. Fragments can be interactively constructed using a visual structure editor. A configurable database browser is designed for inspection and editing of all types of data items. Database statistics is maintained on the number and quality of structures, conformers, and descriptors. Reports can be generated presenting any chosen subset of structures and descriptors into different formats suitable for including into documents. In addition to fixed report formats there is a powerful report template designer module with a visual report template editor to produce customized page layout. The database is compatible at import/export level with SDF, MOL, SMILES and other known formats.

P055 (MEK-1117-802127) Performance, reliability, and improvement of a tissue-specific metabolic simulator.
Start time: 8:00 AM
Mekenyan, O¹, Jones, J², Schmieder, P³, Kotov, S¹, Pavlov, T¹, Dimitrov, S¹, ¹ Laboratory of Mathematical Chemistry, "Prof. Assen Zlatarov" University, Bourgas, Bulgaria² US EPA, ORD, NERL, Ecosystems Research Division, Athens, GA, USA³ US EPA, ORD, NHEERL, Mid-Continent Ecology Division, Duluth, MN, USA
A methodology is described that has been used to build and enhance a simulator for rat liver metabolism providing reliable predictions within a large chemical domain. The tissue metabolism simulator (TIMES) utilizes a heuristic algorithm to generate plausible metabolic maps using a library of measured biotransformations and abiotic reactions. The TIMES simulator prioritizes the order of implementation of metabolic transformation reactions applied to a parent chemical by using probabilities determined from measured data, a significant improvement over earlier approaches. A transformation hierarchy is defined to objectively reproduce a training set of documented metabolic maps thus reducing over-propagation of metabolic maps and enhancing prioritization of generated metabolites according to their stability, amount, solubility, toxicity, etc. The reliability of simulated pathways and metabolites is assessed in comparison to observed metabolism data, thus one can prioritise competing pathways, as well as metabolites, by probability of occurrence and reliability. Reliability estimates are further used to strategically select chemicals for testing to most effectively expand the domain of the simulator. The performance of the simulator is assessed by the extent to which documented maps are reproduced by simulated maps and by the breadth of coverage of chemical structures and their metabolites within the simulator domain. A new iterative procedure has been developed to improve the performance of the simulator while expanding the domain where it is applied with highest reliability. When the simulator is re-trained with new metabolic maps, the system automatically identifies false negative (documented but not predicted) and false positive (predicted but not documented) metabolites. Subsequently, the transformation reaction database is updated by expert knowledge (new transformations are added and/or existing transformations modified) and the simulator performance is enhanced (using the software module SimBuilder) by re-defining the probabilities of the simulator transformation library.

P056 (SER-1117-801357) QSAR evaluation of ER Binding Affinity of Chemicals and Metabolites.
Start time: 8:00 AM
Mekenyan, O¹, Serafimova, R¹, Aladjov, H², Kolanczyk, R², Schmieder, P², Akahori, Y³, Nakai, M³, Jones, J⁴, ¹ Laboratory of Mathematical Chemistry, As. Zlatarov University, Bourgas, Bulgaria² U.S. EPA, Mid-Continent Ecology Division, 6201 Congdon Blvd., 55804, Duluth, MN, USA³ Chemicals Evaluation Research Institute (CERI), 1600 Shimotakano,Sugito-machi,Kitakatsushika-gun, 345-0043, Saitama, Japan⁴ US EPA, ORD, NERL, Ecosystems Research Division, 30605, Athens, GA, USA
Chemicals in commerce are assessed for a variety of potential adverse effects. As governments around the globe strive to meet the challenge of assessing chemicals as endocrine disruptors, the need for hypothesis-driven strategies to prioritize chemicals for testing has risen to the forefront. This study describes quantitative structure-activity relationships (QSAR) to predict chemical ER binding as part of a larger research effort using an iterative process of strategic selection of chemicals for testing, in vitro data generation, in silico model development, etc., to facilitate efficient model evaluation and refinement for user-specified levels of predictive certainty. A multi-dimensional formulation of a COmmon REactivity PAttern (COREPA) modeling approach has been used to investigate chemical binding to the human estrogen receptor (hER). A training set of 656 chemicals includes 500 steroid and environmental chemicals (CERI) and 156 to further explore hER-structure interactions (selected J. Katzenellenbogen references). Analysis of reactivity patterns based on the distance between nucleophilic sites resulted in identification of distinct interaction types: a steroid-like A-B type described by frontier orbital energies and distance between nucleophilic sites with specific charge requirements; an A-C type where steric effects are combined with electronic interactions to modulate binding; and mixed A-B-C. Chemicals are grouped by type, then COREPA models are developed for within specific relative binding affinity ranges of >10%, 10 to 0.1%, and 0.1 to 0.001%. The derived models for each interaction type and affinity range combine specific interatomic distances and a COREPA classification node using < 2 discriminating parameters. The interaction types becoming less distinct in the lowest activity range for each chemicals of each type. A battery of models is presented for pre-screening of parent chemicals, or simulated metabolites (i.e., interfacing of toxic effects predictions with a liver tissue metabolism simulator (TIMES) as described in a companion presentation). [Dr. Aladjov is an NRC Post-doctoral appointee; This abstract does not reflect USEPA policy; CERI acknowledges sponsorship by the Ministry of Economy, Trade and Industry, Japan]

P057 (LAM-1117-748399) New methods in QSAR: Use of structural equation modeling to describe PAH photoinduced toxicity.
Start time: 8:00 AM
Lampi, M¹, Reynolds, M², Dixon, D¹, Greenberg, B¹, ¹ Department of Biology, University of Waterloo, Waterloo, ON, Canada² Department of Psychology, University of Waterloo, Waterloo, ON, Canada
Recent advances in the technological fields related to computing have led to a large increase in the use of sophisticated modeling techniques. Ecotoxicology is no different, and indeed there has been wide application of such technology. A recent introduction to the field has been the use of complex methods, particularly neural network models. A complementary technique that is not widely employed in ecotoxicology is structural equation modeling (SEM). This method encompasses many statistical techniques including regression, factor and path analysis. Structural equation modeling allows for testing of hypotheses regarding relationships between observed and latent (unobserved) variables. Latent variables are theoretical concepts that unite phenomena under a single term, and are expressed in terms of directly measured variables. A novel application of structural equation modeling was used to validate the assumption of a bipartite mechanism for photoinduced toxicity of PAHs that includes photosensitization and photomodification. Two latent variables were created to represent the processes of photosensitization and photomodification, which both contribute to phototoxicity. These were based solely on physicochemical and photodynamic properties of the PAHs. The use of SEM enables the weighting of these properties, and their contribution to each latent variable individually, as well as the contribution of the latent variables to toxicity. Six existing PAH phototoxicity data sets were used and structural equation models were estimated using SEM software. These models accounted for a high amount of variance in six different sets of PAH phototoxicity data from different organisms, while providing insight regarding the contribution of different physicochemical and photodynamic descriptors to toxicity. The flexibility of SEM is evident as the relative contributions of each factor could be determined and compared. This study illustrates the promise for this type of modeling in ecotoxicology, potential future uses which include assessment of synergism, and to predictive models developed for other contaminants.

P058 (BAJ-1117-828648) The significance of transformation in data mining for quantitative structure-toxicity relationships.
Start time: 8:00 AM
Baja, M. Sc., Emmanuel¹, Patungan, Ph.D, Welfredo ², Pizana, Ph.D., Romulo³, Quibuyen, Ph.D., Titos⁴, Hermosilla, Ph.D., Augusto³, ¹ Department of Environmental Medicine, New York University School of Medicine, Tuxedo, NY, USA² School of Statistics, University of the Philippines Diliman, Quezon City, Philippines³ Department of Mathematics, University of the Philippines Diliman, Quezon City, Philippines⁴ Institute of Chemistry, University of the Philippines Diliman, Quezon City, Philippines
A Quantitative Structure-Toxicity Relationship (QSTR) approach based on Multiple Linear Regression (MLR) was used to check if transformations on the response function and descriptors would give a better response-factor relationship. Twenty linear models were used in the study using data sets from forty three carbamate insecticide derivatives. LD₅₀ values of the carbamates were used as the response variable, while the descriptors used were solvent-accessible surface area, volume, mass, van der Waals surface area, partial charges, refractivity, polarizability and octanol-water partition coefficient. The values of the physicochemical and structural descriptors were calculated for the functional groups (R₁ and R₂) attached to the parent compound of the carbamates. Coefficient of determination R², estimate of error variance s², and Fisher test (F-test) p-value were used as criteria for the model selection process. Multiple linear regression analysis showed that the 1/Y transformation of the original data yielded a number of benefits. The transformation helped to normalize the error distribution, stabilized the variance and simplify the response function and descriptors by linearizing a nonlinear dependent-independent relationship.

P059 (AAA-1117-704646) QSARs for the acute toxicity of halogenated benzenes to bacteria in natural waters.
Start time: 8:00 AM
Lu, Guanghua¹, Wang, Chao¹, ¹ College of Environmental Science and Engineering, Hohai University, Nanjing, Jiangsu Province, P. R. China
The concentration values causing 50% inhibition of bacteria growth (24h-IC₅₀) were determined according to the bacterial growth inhibition test method. The energy of the lowest unoccupied molecular orbital and the net charge of carbon atom of 20 halogenated benzenes were calculated by the quantum chemical MOPAC program. The log1/IC₅₀ values ranged from 4.79 for 2,4-dinitrochlorobenzene to 3.65 for chlorobenzene. The quantitative structure-activity relationship model was derived from the toxicity and structural parameters data: log1/IC₅₀ =-0.531(E_LUMO)+1.693(Q_C)+0.163(logP) +3.375. This equation was found to fit well (r² =0.860, s=0.106), and the average percentage error was only 1.98%. Halogenated benzenes and alkyl halogenated benzenes are non-polar narcotics, and have shown hydrophobicity-dependent toxicity, the halogenated phenols and anilines exhibit toxic potency higher than that estimated by their hydrophobicity, whereas 2,4-dinitrochlorobenzene is electrophile with the halogen acting as the leaving group.

P060 (ALA-1117-833478) Chemical selection for endpoint related hierarchical chemical categorization and QSAR model development.
Start time: 8:00 AM
Aladjov, H¹, Kolanczyk, R², Nikolov, N³, Mekenyan, O⁴, Veith, G⁵, Akahori, Y⁶, Nakai, M⁶, ¹ NRC Post-doctoral associate – US EPA, Mid-Continent Ecology Division, Duluth, Minnesota, USA² US EPA, Mid-Continent Ecology Division, Duluth, Minnesota, USA³ Centre of Biomedical Engineering "Ivan Daskalov" – Bulgarian Academy of Science, Sofia, Bulgaria⁴ Laboratory of Mathematical Chemistry, Asen Zlatarov University, Bourgas, Bulgaria⁵ International QSAR Foundation to Reduce Animal Testing, Two Harbors, Minnesota, USA⁶ Chemicals Evaluation Research Institute, Sugito-machi, Japan
While building a single model for estrogen binding affinity may be statistically possible it can be argued that such a model will lack mechanistic interpretation and be extremely sensitive to training set selection. Alternatively, attempts to group chemicals by receptor interaction types allows resultant QSARs to be more specific and reveal structural requirements associated with underlying interaction mechanisms. To achieve this an iterative approach to model development was adopted. Based on available literature four major types of ER interaction were outlined and representative chemicals selected. From this, rules for mechanistic classifications were proposed. When rules were applied to chemical inventories many chemicals remained unclassified, presumably due to incompleteness of the initial rules. New categories were introduced paying attention to first principles and functional groups, and used to parse: a training set designed in-house with a large representation of industrial chemicals (EPA/MED); other existing training sets from CERI and literature sources; and, chemical inventories of interest. By comparing the number of chemicals in each category within training sets and inventories it was possible to outline the relevance of the classes with respect to inventories and assess test data availability for building models. Approaches for selecting and testing chemicals to further investigate structure-activity hypotheses where ambiguous or no information exists were developed. Models built where data permitted were used to further refine and evaluate mechanistic hypotheses and aid further chemical selection. Where particular categories of chemicals are found to be consistent model outliers, new interaction mechanisms can be hypothesized and explored. Structural requirements established in the models are used to further screen inventories and select and test chemicals, eventually stabilizing the number of categories and interaction mechanisms. This iterative process of refining models and categorization rules is continued until a target inventory is adequately covered to the extent needed for regulatory acceptance.

P061 (SCH-1117-829762) In vitro tests, mechanistic hypotheses, and iterative model development to build QSARs for risk assessment.
Start time: 8:00 AM
Schmieder, P¹, Mekenyan, O², Veith, G³, ¹ U.S. EPA, ORD, NHEERL, Mid-Continent Ecology Division, Duluth, MN, USA² Laboratory Mathematical Chemistry, Bourgas As. Zlatarov University, Bourgas, Bulgaria³ International QSAR Foundation to Reduce Animal Testing, Two Harbors, MN, USA
Assessing the risks of chemicals when measured data are not available is a major challenge for regulatory bodies, and the identification of potential endocrine disruptors highlights the need for hypothesis-driven approaches to estimate adverse effects at lower cost. One approach is the development of quantitative structure-activity relationships (QSAR) for major toxicity pathways such as the estrogen signaling pathway. Statistical models of complex receptor binding without classification of binding mechanisms are often unreliable, and yet the elucidation of binding mechanisms for many receptors is still in its infancy. This paper presents an iterative approach for discovery of the structural requirements for distinct mechanisms which involves QSAR-based hypothesis generation, strategic chemical selection for hypothesis testing, QSAR evaluation and improvement of mechanistic classification. The approach is grounded in seeking mechanistic understanding of the underlying interactions, and defining chemical similarity in terms of biological activity. Strategic selection of chemicals for testing is an essential component, given the numerous diverse chemicals for model regulatory applications. The immediate goal is to determine structural requirements for chemical binding to the estrogen receptor, a more complex interaction than previously appreciated. The larger objective is to present a process applicable to recurring issues surrounding determining structural attributes associated with toxicity leading to adverse biological consequence. Determinations must be made with enough specificity to result in reliable predictions but broad applicability to numerous diverse chemicals. The process strives for mechanistic interpretability and transparency, with measures of coverage within inventories to which applied. The presentation introduces key aspects (detailed in posters) integrating information through successive iterations, including: i) in vitro assays (effects, concentrations, metabolites) linked to adverse outcomes; ii) QSAR hypothesis-generation, model refinement; iii) strategic chemical selection in the context of regulatory inventories; and, iv)prioritization of chemicals for the endpoint of concern.[Abstract does not necessarily reflect EPA policy].