Concept drift detection in metabolomics-based predictions under cold-stress conditions in plants

Abstract

In recent years, there has been a significant surge in the use of prediction models. These models typically assume that input data is stationary; however, much of the data we encounter is dynamic in nature. Among the most dynamic data are metabolomics, which change over time. One common challenge with time-varying data is the occurrence of concept drift. Concept drift can degrade the accuracy and reliability of prediction models, making it a key issue to address. Despite its negative impact, concept drift can also be leveraged to uncover hidden confounding factors in the data. The main goal of this study is to detect concept drift in metabolite concentrations across different ecotypes of Arabidopsis thaliana under varying growth conditions. The study is divided into two parts: first, we apply different predictive regressors to metabolomics data, and second, we use these models to detect concept drift. Our focus lies in identifying potential confounding factors that may influence the prediction of relative growth rates in plants.