Date on Master's Thesis/Doctoral Dissertation


Document Type

Master's Thesis

Degree Name



Pharmacology and Toxicology

Committee Chair

Cunningham, Albert R.

Author's Keywords

Structure-activity relationship; Cat-SAR; Endocrine disruptor; Estrogen


Estrogen; Selective estrogen receptor modulators


Xenoestrogens are spread throughout the environment affecting our daily lives and may produce potential toxic effects on human health. The purpose of this study was to develop a mechanistically reliable model capable of identifying xenoestrogens. Our hypothesis was that there are identifiable structural characteristics among a diverse set of estrogen receptor ligands that differentiate estrogenic and nonestrogenic compounds. The model's learning set was developed by collecting compounds from the National Center for Toxicological Research Estrogen Receptor Binding database (NCTRER) . The categorical-SAR (cat-SAR) expert system was used to build the models and perform leave-none-out, leave-one-out, leave-many-out and external validations for model analysis. The values of all validations were between 0.80 and 0.97. Based on several analyses of rational subsets of compounds included in the NCTRER based on potency or chemical structure, it was observed that the developed SAR models predictivity varied across sets. This indicates that variability in the SAR models or the in vitro assay results themselves must be considered when applying SAR models for prediction or mechanistic analyses of estrogen receptor ligands. Fragment analysis was carried out to study the mechanism of estrogen receptor binding, and various important fragments were identified that demonstrate potential structural characteristics important for binding. Furthermore, this led to the discovery that the cat-SAR expert system was able to make a higher percentage of correct predictions on specific classes of xenoestrogen expressing these key functional groups. In conclusion, this estrogen receptor ligand model has good predictive performance and is based on model attributes that are mechanistically sound.