Genomic Expression-based In Silico Predication of Drug Liver Toxicity and Patient's Atherogenic Risk

Cheng, Feng, Department of Molecular Physiology and Biological Physics, University of Virginia
Nakamoto, Robert, Department of Molecular Physiology and Biological Physics, University of Virginia

Gene expression profiling by using microarray is a high throughput gene analysis method that enables scientists to create a global picture of cellular function on a genome-wide scale. In chapter 1, I provide an overview of the most important issues regarding this technique including the basic concept, the statistical algorithm and practical use. In chapter 2, we develop and apply a novel prediction technique for identifying hepatotoxic compounds based on in vitro gene expression profile responses in human liver cells. Using a training set of in vivo rodent experiments, the COXEN algorithm and human liver tissue gene expression profiles, we developed a 32 gene predictor of druginduced liver toxicity. This predictor was evaluated in five independent test sets of in vitro human liver cells and in vivo animal toxicity screenings and consistently demonstrated high prediction performance. In these five test sets, the predicted toxicity scores generated by the model for the samples treated with toxic compounds (or doses) were statistically different (p-value < 0.05) from those treated with low toxic compounds (or doses) or the control group. In three of the five test sets, all toxic and low-toxic groups were perfectly identified by our prediction model (sensitivity and specificity are both 1). The gene signature that we described can also distinguish between toxic and nontoxic doses of the same compounds, suggesting that this approach could also be used to evaluate safe doses of drugs of interest. In chapter 3, we develop a novel in silico technique for the prediction of atherosclerosis risk based on microarray data from white blood cells. By comparing monocytes from 2 familial hypercholesterolemia patientswith normal people and using biological function analysis, we discover 56 atherosclerosis-related blood genomic biomarkers. Using the COXEN algorithm, we further identify three subsets of genes that could be developed for T-lymphocytes, total white blood cells and macrophages from these 56 genes. These models have demonstrated high prediction performance on three independent test sets. This in silico atherogenic risk prediction technique has also shown its utility in clinical applications since it has some advantages over current diagnosis tools and studies.

Note: Abstract extracted from PDF text

PHD (Doctor of Philosophy)
All rights reserved (no additional license for public reuse)
Issued Date: