PEACOCK: a machine learning approach to assess the validity of cell type-specific enhancer-gene regulatory relationships.

The vast majority of disease-associated variants identified in genome-wide association studies map to enhancers, powerful regulatory elements which orchestrate the recruitment of transcriptional complexes to their target genes' promoters to upregulate transcription in a cell type- and timing-dependent manner. These variants have implicated thousands of enhancers in many common genetic diseases, including nearly all cancers. However, the etiology of most of these diseases remains unknown because the regulatory target genes of the vast majority of enhancers are unknown. Thus, identifying the target genes of as many enhancers as possible is crucial for learning how enhancer regulatory activities function and contribute to disease. Based on experimental results curated from scientific publications coupled with machine learning methods, we developed a cell type-specific score predictive of an enhancer targeting a gene. We computed the score genome-wide for every possible cis enhancer-gene pair and validated its predictive ability in four widely used cell lines. Using a pooled final model trained across multiple cell types, all possible gene-enhancer regulatory links in cis (~17 M) were scored and added to the publicly available PEREGRINE database ( www.peregrineproj.org ). These scores provide a quantitative framework for the enhancer-gene regulatory prediction that can be incorporated into downstream statistical analyses.

Dopamine genes and nicotine dependence in treatment-seeking and community smokers.

We utilized a cohort of 828 treatment-seeking self-identified white cigarette smokers (50% female) to rank candidate gene single nucleotide polymorphisms (SNPs) associated with the Fagerström Test for Nicotine Dependence (FTND), a measure of nicotine dependence which assesses quantity of cigarettes smoked and time- and place-dependent characteristics of the respondent's smoking behavior. A total of 1123 SNPs at 55 autosomal candidate genes, nicotinic acetylcholine receptors and genes involved in dopaminergic function, were tested for association to baseline FTND scores adjusted for age, depression, education, sex, and study site. SNP P-values were adjusted for the number of transmission models, the number of SNPs tested per candidate gene, and their intragenic correlation. DRD2, SLC6A3, and NR4A2 SNPs with adjusted P-values <0.10 were considered sufficiently noteworthy to justify further genetic, bioinformatic, and literature analyses. Each independent signal among the top-ranked SNPs accounted for approximately 1% of the FTND variance in this sample. The DRD2 SNP appears to represent a novel association with nicotine dependence. The SLC6A3 SNPs have previously been shown to be associated with SLC6A3 transcription or dopamine transporter density in vitro, in vivo, and ex vivo. Analysis of SLC6A3 and NR4A2 SNPs identified a statistically significant gene-gene interaction (P=0.001), consistent with in vitro evidence that the NR4A2 protein product (NURR1) regulates SLC6A3 transcription. A community cohort of N=175 multiplex ever-smoking pedigrees (N=423 ever smokers) provided nominal evidence for association with the FTND at these top ranked SNPs, uncorrected for multiple comparisons.

A systems biology network model for genetic association studies of nicotine addiction and treatment.

OBJECTIVE: Interpreting genome-scale genetic association data, particularly for complex diseases and phenotypes, requires extensive use of prior knowledge across a broad range of potential biological and environmental influences, spanning many scientific subdisciplines. We suggest that known or hypothesized disease risk factors, and causal mechanisms, can be represented using an ontology, a computational specification of a set of concepts and the relations between them. METHODS: We have integrated the expertise of multiple investigators in nicotine pharmacokinetics and pharmacodynamics, nicotine dependence, and clinical smoking cessation outcomes, and represented this knowledge in an ontology-based network model. Our model spans multiple scales, from molecules, genes and cellular pathways, to complex behavioral phenotypes and even environmental factors. To leverage previous and ongoing work in the field of ontology development, we adopt, expand upon and relate elements from existing ontologies whenever possible. RESULTS: We discuss several applications of our ontology: to support interdisciplinary research by graphically representing a complex scientific theory, to facilitate meta-analysis across different studies, to highlight potential interactions, and to support statistical analysis and causal modeling. We demonstrate that our ontology can focus hypothesis testing on areas supported by current theory. CONCLUSION: We describe how an ontology-based computational representation can be applied to disease risk factors and mechanisms, enabling the use of prior knowledge in large-scale genetic association studies in general. In specific, we have developed an initial Smoking Behavior Risk Ontology to support studies related to the pharmacogenetics of nicotine addiction and treatment.