Introducing forest plot in representing the relative abundance analysis using parametric and non-parametric models in type 1 diabetes (T1D) human infant gut microbiome

Background In this project, different statistical methodologies were compared to study abundances of bacterial species. Research on this topic is relevant due to the special distributional properties of microbiome data, such as zero inflation, overdispersion, and correction for library size. Additionally, the forest plot was introduced as a way to present the fold changes in abundances. Methods Kostic (Cell Host&Microbe, 2015) studied the relationship between human gut microbiome dynamics throughout infancy and T1D. Based on this study, five different methodologies were compared to study the fold change between T1D cases versus non-cases. This included negative binomial and zero-inflated negative binomial with one and two dispersion parameters, and the Wilcoxon test. Results Out of 2146 OTUs (Operational Taxonomy Units), 40 had significant differences between the disease groups in both models with disease-specific dispersion parameters, while there were no differences found in Wilcoxon tests. Conclusion There is a striking difference of the number of significances found between the five methods. Wilcoxon test may not have found any significances due to the lack of power. The difference between the parametric methods could relate to different overdispersions in different groups. Venn diagrams and forest plots are a good way to illustrate the results.