Content area
Abstract
This dissertation is motivated by my perception that statistical geneticists underutilize causal models. To motivate statistical geneticists to use causal models when investigating the causes of complex genetic disease I have (1) illustrated (a) causal models for complex disease, (b) the distinction between causal and statistical concepts, and (c) the utility of a clear concept of causation for statistical genetics research; (2) determined two-locus penetrance matrices for causal models of gene-gene interaction, causal genetic heterogeneity, and a single genetic cause of disease, and contrasted those models with the more well-known additive, heterogeneity, and multiplicative penetrance models; and (3) used the causal model-based two-locus penetrances to evaluate whether familial aggregation data can be used to infer the number of genes causing disease or whether they interact or act independently to cause disease.
While statistical geneticists often employ an additive penetrance model to describe two-locus penetrance for two genes acting independently to cause disease (causal genetic heterogeneity), and a multiplicative penetrance model to describe two-locus penetrance for two genes acting together (causal gene-gene interaction) to cause disease, I found that these statistical models do not describe the causal mechanisms as intended. An additive penetrance model describes causal genetic heterogeneity only if the genetic causes act in concert (interact) to cause disease in some individuals in the population and even then under highly constrained conditions. A multiplicative penetrance model describes causal gene-gene interaction only when either (1) phenocopies are absent and exactly one of the interacting genes also causes disease through an independent mechanism, or(2) causal genetic heterogeneity is absent and highly constrained conditions prevail. A genetic heterogeneity penetrance model, as described by Risch, describes a disease caused by both genetic heterogeneity and gene-gene interaction if one of the genotypes demonstrates complete penetrance. When neither genotype is fully penetrant; however, the genetic heterogeneity model describes the distribution of disease and genotypes in the population only in the absence of underlying gene-gene interaction.
Using a two-locus penetrance model based on causal principles, I found that familial aggregation patterns can, in some cases, distinguish whether a disease is caused by two independently acting genes, or by two interacting genes, and provide information about the number of interacting genes. More broadly, I demonstrated the applicability and utility of causal models for statistical genetics research.
