200 likes | 319 Views
Toxicological Knowledge Base (a Definition).
E N D
Toxicological Knowledge Base (a Definition) “An in computeroaggregated set of the most germane literature citations and biological activity datasets, together with computational models which correlate those activities with chemical structure, and, ultimately, models for risk assessment.” Response Structure Dose Correlation Activity
A circular process provides training sets DESIGNED to produce robust, living predictive models Data in scientific papers Validation: 1. Cross-validation 2. Testset challenge Develop Models EDKB Data validation and selection Diversity analysis Training set design New Data Needs & Research Hypothesis NCTR/EPA Experiments
EDKB OBJECTIVES • Develop and validate predictive computational models for ER and AR by: • developing appropriate data sets to train models • examining a wide variety of models for performance and ease of use • validating models by: • internal validation • external validation from non-randomly selected chemicals • external validation from randomly selected chemicals from the 58,000 chemical EPSD
Dataset Criteria for Development of SAR models • Develop training data set by design for • structurally diverse chemical structures • a wide range of binding values (RBAs)
Status of ER Models • 238 chemicals assayed (duplicate tubes; two replicates) over a six log RBA range • Techniques developed for identification of potential binders by EDKB team • Filters (Phase I) and 11 categorical SAR models (Phase II) finished • Models validated against literature datasets
Status of ER Models (2) • Models biased toward false positives to minimize false negatives • RBA predictions made for all 58,000 chemicals “in play” • Handout of ER model publications • ~ First two years-used literature RBAs • Subsequently used NCTR dataset
Androgen Receptor Competitive Binding Assay Two assays evaluated: • Ventral prostate from castrated adults • Could not obtain reproducible results • PanVera androgen binding domain • Good saturation and competitive binding results • Assay subsequently validated for use • Less expensive than ventral prostate • Uses no animals
Validation of PanVera Assay • PanVera protein was diluted to several low concentrations and saturation assays were conducted with [3H]-R1881. • Protein concentrations selected were in the range of the reported KD. • A protein concentration in the stable KD range was used for subsequent competitive binding assays.
Status of AR Assay Up to May 9, 2001, 134 compounds were tested. To date,196 compounds have been tested Number of Chemicals
Activity Distribution for AR Number of Chemicals
De novo external validation of assays and models • EPA funded $850K contract with Batelle Northwest for validation and comparison performance of NCTR and EPA assays and models • Assays methods include: • uterine cytosol for ER (EPA and NCTR assays are similar) • prostate assay for AR (EPA) • PanVera assay for AR (NCTR)
EDKB EXTERNAL FUNDING • SOURCE AMOUNT ($) • FDA OWH 1 185,000 • FDA OWH 2 325,000 • CMA CRADA 2 420,000 • EPA GRANT 1 425,000 • EPA IAG 32,000,000 • TOTAL 3,355,000 • 1 Experimental collaboration with U Missouri • 2 Two separate awards • 3 $500,000 for FY 2001; $1,500,000 for FY 2002-2005