A novel data fusion strategy of LC-MS and NMR technologies using random forest model for emodin hepatotoxic metabolomics research

Data fusion is a process of integrating data matrices from different detection sources into a comprehensive model, which can more comprehensively mine the characteristics of samples and provide the possibility of obtaining more accurate classification. In our previous study, liquid chromatography-mass spectrometry (LC-MS) and nuclear magnetic resonance (NMR) techniques were used to investigate liver metabolomics before and after emodin administration. However, the classification effect between groups of emodin with different durations of administration was not ideal. In this study, principal component analysis, partial least squares discrimination analysis, support vector machine, k-nearest neighbor, neural network, decision tree, and random forest (RF) algorithms were applied for building low-level (LLDF) and mid-level data fusion (MLDF) models. The model classification effects of both the single data set and LLDF were poor, while the separation effect of the MLDF model was significantly improved, among which an RF model established after combining the features selected by the RF model (RF-RF) had the best classification effect. This was the first study to apply LC-MS and NMR data fusion techniques to emodin hepatotoxic metabolomics research, to explore the differential metabolites at a more comprehensive and deeper level, and to provide a research basis for subsequent mechanism studies.