Metabolomics Data Encoded into Images for AI-Based Clinical Diagnosis

Published on: 

A new approach developed by researchers called MetImage can help decode liquid chromatography–mass spectrometry (LC–MS)-based untargeted metabolomics data into multichannel digital images, which is a new approach for using whole metabolome profiles for AI-based clinical applications.

Metabolomics is a field of study that focuses on identifying and quantifying small molecules, or metabolites, in biological samples. Liquid chromatography–mass spectrometry (LC–MS) is a widely used technique for metabolomics analysis. However, the analysis of LC–MS-based untargeted metabolomics data can be technically challenging. This is particularly true for clinical diagnosis. A recent study published in Analytical Chemistry has introduced an innovative approach called MetImage that encodes LC–MS-based untargeted metabolomics data into multichannel digital images for AI-based clinical diagnosis (1).

The study was conducted by researchers from the Shanghai Institute of Organic Chemistry at the Chinese Academy of Sciences, led by corresponding author Zheng-Jiang Zhu. MetImage is designed to generate images that represent comprehensive metabolome profiles. These metabolome profiles can then be used for developing deep learning-based AI models toward clinical diagnosis. The researchers demonstrated the application of MetImage for clinical screening of esophageal squamous cell carcinoma (ESCC) in a clinical cohort of 1104 participants.

A convolutional neuronal network-based AI model was trained to distinguish ESCC screening positive and negative subjects using their serum metabolomics data. The AI model showed superior performance, with sensitivity at 85%, specificity at 92%, and area under curve at 0.95, which were validated in an independent testing cohort of 442 individuals. The encoded images reserved the characteristics of mass spectra from the raw LC–MS data, enabling the identification of metabolites in key image features.


MetImage provides a unique solution to overcome the technical challenges of LC–MS-based untargeted metabolomics data analysis for AI-based clinical diagnosis. It not only allows for the utilization of whole metabolome profiles, but it also offers improved interpretability.

In summary, the study demonstrated the potential of MetImage for clinical applications and paves the way for further development of AI-based diagnostic tools for various diseases.


(1) Wang, H.; Yin, Y.; Zhu, Z.-J.Encoding LC–MS-Based Untargeted Metabolomics Data into Images toward AI-Based Clinical Diagnosis. Anal. Chem. 2023, ASAP. DOI: 10.1021/acs.analchem.2c05079