UCLA

This dissertation develops computational methods for analyzing large-scale genomic and epigenomic datasets. We developed a supervised machine learning approach to predict non-exonic evolutionarily conserved regions in the human genome based on vast amount of functional genomics data. The resulting probabilistic predictions provide a resource for prioritizing functionally important regulatory regions in the human genome. We also developed a method for identifying from large-scale gene expression datasets genes that are differentially expressed in both blood and brain from 12 vervet monkeys, which we used to identify 29 transcripts whose expression is variable between individuals and heritable. Additionally, we developed a method using a global search optimization algorithm to successfully improve a model of human thyroid hormone regulation dynamics leading to a better fit of data for thyrotoxicosis. Together, these three approaches have the potential to impact the understanding and eventual treatment of disease.