Association mapping of traits over time using Gaussian processes
The goal of genetic association studies is to identify the causal genetic regulators that explain the variation of the phenome across individuals or samples. While for univariate, multivariate and groups of phenotypic traits, various computational approaches have been proposed, suitable methods for mapping phenotypes that are recorded over time yet remain to be explored. In this work we propose a Gaussian-process based association model that exploits the temporal structure of time series of phenotypes. Our model is able to identify predictive regulators for time series recordings by jointly clustering the observed time series and mapping the uncovered cluster structure to genetic loci. On synthetic datasets, our GP model detects true genetic regulators with higher accuracy than two state-of-the-art comparison methods. In experiments on clinical data, our GP association model allows us to accurately predict antidepressant treatment response over time based on the genetic state of a patient.