Marta Kordalewska, Department Of Biopharmaceutics And Pharmacodynamics, Medical University Of Gdańsk, Gdańsk, Poland (marta.kordalewska@gumed.edu.pl)

ABSTRACT

The key element of every data analysis in metabolomics should be the identification of the distribution of the generated data. This is important to consider if our goals are to understand the metabolomics data and to ensure that we can generate unbiased coefficients and mean values that generate reasonable predictions.

Metabolomics data generated from GC-MS or LC-MS experiments are usually non-normally distributed. Identifying metabolites with binomial distribution (recognized by the presence of two modes, each with a characteristic peak) is an important task in metabolomics data analysis as this distribution may occur either naturally from group separation, or may indicate problems with identification. Regardless of its origin a useful tool for the analysis is binomial regression which has the flexibility to fit various distributions.

The main aim of this study is to illustrate how to simply identify the structure of metabolomics data exemplified on 9 randomly selected metabolites using basic plots and how to model bimodally distributed data using a binomial regression (Bayesian Regression Models using 'Stan' – brms package). Also, “ranking and selection” was performed to identify metabolites with the greatest contribution for group separation.

ACKNOWLEDGEMENTS

This project was supported by the National Science Centre allocated on the basis of the decisions numbers 2015/19/N/NZ7/03397.

---

Abstract Reference & Short Personal Biography of Presenting Author

Marta Kordalewska graduated in Pharmaceutical Sciences at the Faculty of Pharmacy at Medical University of Gdańsk in 2013. Since then, she is a PhD student at the Department of Biopharmaceutics and Pharmacodynamics at Medical Uiversity of Gdańsk. Her research topic focuses on application of separation techniques (LC-MS, GC-MS, CE-MS) and bioinformatics in metabolomics.