Computational Modeling of Dose Response Relationships for Steroid Hormone Receptor-mediated Gene Expression and Prediction of Androgen Response Element
No Thumbnail Available
Files
Date
2008-03-05
Authors
Journal Title
Series/Report No.
Journal ISSN
Volume Title
Publisher
Abstract
Steroid hormone receptors are critical targets of both synthetic drugs used in hormone therapy and environmental endocrine active chemicals (EAC). Gene expression mediated by steroid hormone receptors was found to exhibit a non-monotonic dose response relationship. To further investigate this relationship, an ordinary differential equation-based computational model was formulated to examine the effect of EACs that display non-monotinic, rather than the typical monotonic, dose-response behaviors. Where the agonist ligand is an agonist, a U-shaped dose-response appears as a consequence of the inherently nonlinear process of receptor homodimerization. A higher degree of U-shaped dose-response curve modulation is effected by mixed-ligand heterodimers formed between endogenous and exogenous ligand-bound monomers. A novel mechanism for non-monotonic, particularly U-shaped, dose-response behaviors observed with specific steroid homologs is provided through this work. This mechanism will help in not only understanding how selective steroid receptor modulators work, but also in the improvement of risk assessment for EACs.
Another focal point in this research is on the statistical approaches to the identification of androgen response elements (ARE). The regulation of gene expression is largely influenced by the behavior of DNA-binding transcription factors. Several computational methods have demonstrated their ability to predict transcription factor binding sites (TFBS) in the gene promoter regions. Namely, Support Vector Machine (SVM), Hidden Markov Model (HMM), and Random Forest (RF) all summarize sequence patterns of experimentally determined TFBS. In order to strengthen the prediction of putative AREs in the human genome, three statistical methods were explored, whose cross-validation results indicated that they all provided good sensitivity and specificity in identifying AREs, with an accuracy of at least 80%. It is the first time HMM, SVM, and RF have all been applied to the construction of ARE prediction models.
As a complement to the first two topics, an elucidation of androgen receptor-dependent gene regulatory networks was pursued. Understanding the underlying mechanism and dynamics of androgen receptor-regulated gene networks requires knowing the direct target genes whose expression levels are modulated by the androgen signaling pathway. Through the inspiration of the Arabidopsis transcription network, a systematic analysis of human genome gene upstream regions was undertaken with the goal of identifying potential androgen response regulatory elements through the occurrences of regular expression patterns. A number of interactions have been suggested between the AR and other target gene transcription factors as a result of the transcription network and sequence analysis of the functional targets.
Description
Keywords
steroid hormone receptor, androgen receptor, random forests, support vector machine, , hidden markov model
Citation
Degree
PhD
Discipline
Bioinformatics
