Exploratory analysis of metabolic changes using mass spectrometry data and graph embeddings.

Area of Science:

• Biochemistry
• Bioinformatics
• Data Science

Background:

• Mass spectrometry (MS)-based metabolomics is crucial for understanding metabolic networks.
• Untargeted metabolomics studies face challenges in data mining due to large datasets and the need to identify subtle changes.
• Traditional statistical methods can overfilter data, potentially removing relevant information and reducing the identification of true metabolic changes.

Purpose of the Study:

• To address the limitations of traditional data mining techniques in MS-based metabolomics.
• To introduce a novel deep learning approach for analyzing untargeted metabolomic data.
• To improve the accuracy and efficiency of identifying metabolic changes from complex MS datasets.

Main Methods:

• Development of GEMNA (Graph Embedding-based Metabolomics Network Analysis), a novel deep learning approach.
• Utilizing graph neural networks (GNNs) for node and edge embeddings.
• Incorporating anomaly detection algorithms for data analysis.
• Application to an untargeted volatile metabolomics study on Mentos candy.

Main Results:

• GEMNA demonstrated superior data clustering compared to traditional tools.
• The novel deep learning approach achieved better data organization and reduced variability.
• Quantitative improvements in data analysis were observed, with GEMNA showing [Formula: see text] compared to [Formula: see text] for the traditional approach.

Conclusions:

• GEMNA offers a powerful and improved alternative for mining untargeted MS-based metabolomic data.
• Deep learning, particularly graph embeddings and anomaly detection, can overcome the limitations of traditional statistical methods.
• This approach enhances the ability to identify and interpret metabolic changes in complex biological systems.