Applied Statistics and Data Science Seminar on Monday, August 21, 2023

Title: Hierarchical structural component models for pathway analysis of longitudinal categorical phenotypes

Presenter: 

Md. Kamruzzaman, PhD

Associate Professor, Jagannath University, Email: kzaman1@isrt.ac.bd

Abstract:

Several statistical methods for pathway analysis have been developed to test the association between pathways and phenotypes of interest. Since pathways are highly correlated, a hierarchical structural component models (HisCoM) was developed to analyze all pathways in a single model and take into consideration their correlation. HisCoM was originally developed to analyze a single phenotype using only one measurement per individual. Later, it was extended to analyze multiple phenotypes (HisCoM-multi) and longitudinal phenotypes (HisCoM-GEE). These methods have been used to analyze continuous, counts, and binary phenotypes from cross-sectional, clustered, and longitudinal studies. In this study, we propose a hierarchical structural component model for pathway analysis of longitudinal categorical phenotypes (HisCoM-RCateg). HisCoM-RCateg is proposed by combining the hierarchical structural component model and generalized estimating equations for correlated categorical phenotypes. HisCoM-RCateg accounts for the biological hierarchy of all biomarkers and pathways into a single model. In the simulation, the proposed HisCoM-RCateg appeared to have high power than other existing methods and effectively controlled type I error for longitudinal multinomial phenotypes. To demonstrate the performance, we also applied HiscoM-RCateg to two distinct types of longitudinal omics data, namely the metabolite dataset and the metagenome dataset.