Leveraging AI and Machine Learning ML for Untargeted Metabolomics and Exposomics: Advances, Challenges, and Future Directions (2024)

AI and ML in Untargeted Metabolomics and Exposomics:

Metabolomics employs a high-throughput approach to measure a variety of metabolites and small molecules in biological samples, providing crucial insights into human health and disease. One application, untargeted metabolomics, allows for an unbiased global analysis of the metabolome, identifying key metabolites that contribute to or indicate health conditions. Recent advances in AI and ML have significantly enhanced untargeted metabolomics workflows, especially in the context of high-resolution mass spectrometry (HRMS) exosomes. This emerging field detects endogenous metabolites and exogenous chemicals in human tissue, linking environmental exposures with disease outcomes. AI and ML applications have improved data quality, rigor, detection, and chemical identification, facilitating major disease screening and diagnosis findings.

Metabolism, the body’s process of producing essential metabolites, includes catabolism (breakdown of molecules for energy) and anabolism (synthesis of compounds needed by cells). Metabolomics captures endogenous metabolites and signaling molecules involved in gene expression, protein function, and enzyme activity. Targeted metabolomics measures specific metabolites, while untargeted metabolomics provides a broader, semi-quantitative analysis of thousands of small molecules. This holistic approach, termed exposomics, incorporates environmental exposures, diet, lifestyle, and psychosocial factors, revealing their impact on health. Despite the vast unknowns in the human exposome, AI and ML are advancing the detection and analysis of these complex datasets, improving the understanding of chemical exposure and its effects on human health.

Untargeted Metabolomics Workflow:

For analyzing biological matrices such as serum, plasma, or urine, the untargeted metabolomics workflow typically involves the separation of complex mixtures using LC or GC column chromatography, followed by HRMS detection and measurement. The process includes sample preparation, data acquisition, pre-and post-processing, data analysis, and chemical identification. Metabolites and chemicals are extracted using organic solvents and analyzed through HILIC or reverse-phase chromatography for LC or derivatized for GC analysis. The HRMS generates data in three dimensions: mass-to-charge ratio, retention time, and abundance. AI and ML tools play crucial roles in data processing, feature selection, and chemical identification, enhancing the analysis of metabolomics data and its biological interpretation.

Data Processing in Untargeted Metabolomics:

Metabolomics raw data are complex due to linear and non-linear interactions among metabolites and challenges with mass spectrometry data structure. Pre-processing is crucial for translating 3-D data from LC-MS into a 2D aligned peak table, which is necessary for downstream analysis. Algorithms like XCMS, MZmine, and MS-Dial are used for pre-processing, but only some methods are universally accepted. Recent developments include quality control measures and new peak-picking algorithms, such as CPC and Finnee, which enhance peak selection. Machine learning tools like WiPP, MetaClean, Peakonly, NeatMS, NPFimg, and EVA promise to improve data processing accuracy and reliability.

AI and ML in Biomarker Discovery:

Traditional univariate and multivariate models perform multiple hypothesis tests to identify metabolite features associated with phenotypes but need help with the correlated structure of metabolomics data. AI and ML methods address these limitations by building and testing models directly on the data, uncovering relationships between phenotypes, exposures, and diseases. Tools like LASSO, PCA, HCA, SOMs, PLS-DA, RF, and newer methods like ANNs and DL have successfully identified significant biomarkers and metabolite signatures. AI and ML have been used to detect diseases like NAFLD, COVID-19, Alzheimer’s, and depression, demonstrating their potential in metabolomic research.

Metabolite Identification in Biomarker Discovery:

Metabolite identification is vital in biomarker discovery, requiring the annotation of selected peaks using metabolite databases and spectral libraries like GNPS, Metlin, and the Human Metabolome Database. This process involves matching m/z and MS/MS fragmentation data to confirm metabolites. Despite available databases, spectral matching rates for specialized chemicals still need to be higher. Advances in cognitive metabolomics using ML and NLP and in silico tools like CSI: FingerID and CFM-ID are improving identification accuracy. Expanding spectral libraries and developing new annotation tools are crucial for better identification and understanding of both endogenous and exogenous chemicals.

Advances in Untargeted Chemical Analysis:

Advances in untargeted chemical analysis and AI/ML tools have significantly reduced costs, enabling large-scale studies. AI/ML aids in data extraction, mining, and annotation, which is crucial in biomarker discovery. The primary challenge remains annotating unknown metabolites essential for biological interpretation. Efforts focus on developing experimental databases and AI/ML models to enhance metabolite identification. However, current algorithms often miss low-concentration chemicals, indicating a need for improved ML classifiers. Integrating a biology-driven approach with measurement-based methods may uncover unknown chemicals affecting health, catalyzing discoveries in exosomes and precision health.

Check out the Paper. All credit for this research goes to the researchers of this project. Also,don’t forget to follow us onTwitter and join ourTelegram Channel andLinkedIn Group. If you like our work, you will love ournewsletter..

Don’t Forget to join our47k+ ML SubReddit

Find Upcoming AI Webinars here