EigenRF: an improved metabolomics normalization method with scores for reproducibility evaluation on importance rankings of differential metabolites

Abstract

Screening differential metabolites is of great significance in biomarker discovery in metabolomics research. However, it is susceptible to unwanted variations introduced during experiments. Previous normalization methods have improved the accuracy of inter-group classification by eliminating systematic errors. Nonetheless, the classification ability of differential metabolites obtained through these methods still requires further enhancement, and the reproducibility evaluation on importance rankings of differential metabolites is often disregarded. The EigenRF algorithm was developed as an improvement over the previous metabolomics normalization method referred to as EigenMS, which aims to normalize metabolomics data. Furthermore, scoring metrics, including the local consistency (LC) and overall difference (OD) scores, were introduced to evaluate the reproducibility of importance rankings of differential metabolites from a dual perspective. After conducting validation on three publicly accessible datasets, the EigenRF method has demonstrated enhanced classification ability of differential metabolites as well as improved reproducibility. In summary, EigenRF enhances the reliability of differential metabolites in metabolomics research, benefiting the further exploration of molecular mechanisms underlying biological alterations in complex matrices. The EigenRF algorithm was implemented in an R package: https://www.github.com/YangHuaLab/EigenRF.