Bayesian Matrix Completion for Hypothesis Testing

The United States Environmental Protection Agency (EPA) screens thousands of chemicals primarily to differentiate those that are active vs inactive for different types of biological endpoints. However, it is not feasible to test all possible combinations of chemicals, assay endpoints, and concentrations, resulting in a majority of missing combinations. Our goal is to derive posterior probabilities of activity for each chemical by assay endpoint combination. Therefore, we are faced with a task of matrix completion in the context of hypothesis testing for sparse functional data. We propose a Bayesian hierarchical framework, which borrows information across different chemicals and assay endpoints. Our model predicts bioactivity profiles of whether the dose-response curve is constant or not, using low-dimensional latent attributes of chemicals and of assay endpoints. This framework facilitates out-of-sample prediction of bioactivity potential for new chemicals not yet tested, while capturing heteroscedastic residuals. We demonstrate the performance via extensive simulation studies and an application to data from the EPA's ToxCast/Tox21 program. Our approach allows more realistic and stable estimation of potential toxicity as shown for two disease outcomes: neurodevelopmental disorders and obesity.