Conditional decomposition diagnostics for regression analysis of zero-inflated and left-censored data

A Review on Differential Abundance Analysis Methods for Mass Spectrometry-Based Metabolomic Data

This review presents an overview of the statistical methods on differential abundance (DA) analysis for mass spectrometry (MS)-based metabolomic data. MS has been widely used for metabolomic abundance profiling in biological samples. The high-throughput data produced by MS often contain a large fraction of zero values caused by the absence of certain metabolites and the technical detection limits of MS. Various statistical methods have been developed to characterize the zero-inflated metabolomic data and perform DA analysis, ranging from simple tests to more complex models including parametric, semi-parametric, and non-parametric approaches. In this article, we discuss and compare DA analysis methods regarding their assumptions and statistical modeling techniques.

EM Estimation for Zero- and k-Inflated Poisson Regression Model

Count data with excessive zeros are ubiquitous in healthcare, medical, and scientific studies. There are numerous articles that show how to fit Poisson and other models which account for the excessive zeros. However, in many situations, besides zero, the frequency of another count k tends to be higher in the data. The zero- and k-inflated Poisson distribution model (ZkIP) is appropriate in such situations The ZkIP distribution essentially is a mixture distribution of Poisson and degenerate distributions at points zero and k. In this article, we study the fundamental properties of this mixture distribution. Using stochastic representation, we provide details for obtaining parameter estimates of the ZkIP regression model using the Expectation–Maximization (EM) algorithm for a given data. We derive the standard errors of the EM estimates by computing the complete, missing, and observed data information matrices. We present the analysis of two real-life data using the methods outlined in the paper.

Two-group comparisons of zero-inflated intensity values: the choice of test statistic matters

MOTIVATION

A special characteristic of data from molecular biology is the frequent occurrence of zero intensity values which can arise either by true absence of a compound or by a signal that is below a technical limit of detection.

RESULTS

While so-called two-part tests compare mixture distributions between groups, one-part tests treat the zero-inflated distributions as left-censored. The left-inflated mixture model combines these two approaches. Both types of distributional assumptions and combinations of both are considered in a simulation study to compare power and estimation of log fold change. We discuss issues of application using an example from peptidomics.The considered tests generally perform best in scenarios satisfying their respective distributional assumptions. In the absence of distributional assumptions, the two-part Wilcoxon test or the empirical likelihood ratio test is recommended. Assuming a log-normal subdistribution the left-inflated mixture model provides estimates for the proportions of the two considered types of zero intensities.

AVAILABILITY

R code is available at http://cemsiis.meduniwien.ac.at/en/kb/science-research/software/