False Discovery Rate Control Under Rounding of P-Values
Dr Hien Nguyen
Location : CRIUGM (http://www.criugm.qc.ca/en/contact.html), room M6804
Date: December 13th, 13h-14h
The seminar will be presented in English
Dr Hien Nguyen is going to give special seminar of Unité de Neuroimagerie Fonctionnelle (UNF) this coming Wednesday. Dr Nguyen is a Lecturer and Australia Research Council DECRA Research Fellow at La Trobe University in Melbourne Australia. His research currently focuses on the development of Big Data methodologies that are deployable on small-scale computing infrastructures, and deep learning and neural networks for applications in personalised medicine.
The mitigation of false positives is an important issue when conducting multiple hypothesis testing. The most popular paradigm for false positives mitigation is via the control of the false discovery rate (FDR). We present a method for FDR control that is applicable in cases where only p-values are available, and when those p-values are potentially equal to zero or one. Our method is based on an empirical-based paradigm where the Probit transformation of the p-values (called the z-scores) are modeled as a two-component mixture of normal distributions. Due to the rounding of the p-values, the usual approach for fitting mixture models cannot be applied. We instead use a binned data technique, which can be proved to consistently estimate the z-score distribution, even when the data are correlated. A simulation study shows that our methodology is competitive with popular alternatives, especially when data are correlated. We demonstrate the applicability of our methodology in practice via a brain imaging study of mice.