Structure-revealing data fusion model with applications in metabolomics

Area of Science:

• Metabolomics
• Bioinformatics
• Data Science

Background:

• Joint analysis of multi-source data enhances knowledge discovery in various fields.
• Metabolomics often uses complementary techniques like Liquid Chromatography-Mass Spectrometry (LC-MS) and Nuclear Magnetic Resonance (NMR) spectroscopy.
• Identifying common and unique structures across these datasets is a significant challenge in data fusion.

Purpose of the Study:

• To propose a novel unsupervised data fusion model.
• To address the challenge of identifying common and individual structures across multiple data sets.
• To apply the model to metabolomics data for disease-related biomarker discovery.

Main Methods:

• Developed a data fusion model based on joint factorization of matrices and higher-order tensors.
• The model is unsupervised, automatically revealing common and individual components.
• Applied the model to analyze combined fluorescence and NMR metabolomics data from plasma samples.

Main Results:

• The proposed model successfully identified common and individual components within the metabolomics data.
• Demonstrated promising results in the joint analysis of fluorescence and NMR data.
• Achieved effective separation of colorectal cancer patients from control groups using the fused data.

Conclusions:

• The novel unsupervised data fusion model effectively integrates complementary metabolomics data.
• The joint factorization approach facilitates the discovery of shared and unique biological structures.
• This method holds potential for improving disease diagnostics through enhanced metabolomics data analysis.