A0493
Title: Randomized quantile residuals for diagnosing zero-inflated models with applications to microbiome count data
Authors: Cindy Feng - Dalhousie University (Canada) [presenting]
Abstract: Zero-inflated generalized linear models, particularly zero-inflated negative binomial models, are commonly used for differential abundance analysis of microbiome and other sequencing count data. When estimating the false discovery rate (FDR), it's assumed that p-values follow a uniform distribution under the null hypothesis. Ensuring that the chosen model adequately fits the count data is crucial to control FDR and avoid excess false discoveries. Model checking is, therefore, essential in this analysis. The randomized quantile residual (RQR) method has been shown to effectively diagnose the count regression models. However, its performance in diagnosing zero-inflated generalized linear mixed models (GLMMs) for sequencing count data has not been extensively studied. Large-scale simulation studies are conducted to assess the performance of RQRs for zero-inflated GLMMs. The simulations demonstrate that the type I error rates of the goodness-of-fit tests with RQRs closely match the nominal level. Scatter plots and QQ plots of RQRs are valuable for distinguishing between good and bad models. RQRs are also applied to diagnose six GLMMs in a real microbiome dataset, finding that RQR is an excellent tool for diagnosing GLMMs for zero-inflated count data, particularly in microbiome studies.
