Evaluating variance in liability explained by individual genetic variants and relationship to individualized risk prediction

Abstract

An increasing number of susceptibility genes have been identified for complex diseases in recent years. However, how much the candidate genes discovered to date could explain the total genetic component of a disease is unknown. We developed a statistical framework to address this problem focusing on dichotomous disease traits and applied it to real examples of complex diseases. The genes were mainly selected based on results from meta-analyses of association studies. The total variance contributed by known candidate genes for each disease is generally not high, implying that a substantial part of heritability for most complex diseases remains unexplained. We also extended our model to deal with multi-allelic loci, haplotypes as well as gene-gene and gene-environmental interactions. In addition, we derived methods for calculating the variance explained for continuous predictor variables. We further found that the variance explained is closely related to the ability of risk prediction for individuals. We developed a methodology to quantify the absolute disease risk from liability measures. Specificity and sensitivity could be calculated for every cutoff of the absolute disease risk, hence the receiver operating characteristic curves and areas under the curve may be computed. Finally, we developed an approach to incorporate family history of the individual into known genetic factors when predicting disease risks.