|
Abstract:
|
Predicting an individual’s phenotypic value from their genetic data is a goal and current research area for the field of genetics. This would not only serve as a public heath tool but could provide researchers with an opportunity to increase power for their analyses by increasing the sample size. The primary difficulty in creating such a model lies in the number of loci that contribute to a disease compared to the sample sizes used in training the model. Several penalized regression, Bayesian regression, and non-linear prediction methods have been developed to account for such limitations, but there is no robust method that performs best in every scenario. This presentation will address the main methods used for phenotype prediction and offer a comparison between the methods among a variety of genetic architecture assumptions for a phenotype, as well as discussing limitations of such methods and possible future areas of research.
|
