Deep learning meets metabolomics: a methodological perspective

Abstract

Deep learning (DL), an emerging area of investigation in the fields of machine learning and artificial intelligence, has markedly advanced over the past years. DL techniques are being applied to assist medical professionals and researchers in improving clinical diagnosis, disease prediction and drug discovery. It is expected that DL will help to provide actionable knowledge from a variety of ‘big data’, including metabolomics data. In this review, we discuss the applicability of DL to metabolomics, while presenting and discussing several examples from recent research. We emphasize the use of DL in tackling bottlenecks in metabolomics data acquisition, processing, metabolite identification, as well as in metabolic phenotyping and biomarker discovery. Finally, we discuss how DL is used in genome-scale metabolic modelling and in interpretation of metabolomics data. The DL-based approaches discussed here may assist computational biologists with the integration, prediction and drawing of statistical inference about biological outcomes, based on metabolomics data.

Metabolic Signatures of the Exposome—Quantifying the Impact of Exposure to Environmental Chemicals on Human Health

“…ANNs and associated deep learning methods currently suffer from issues of interpretability and are difficult to apply to the development of simpler clinical tests; other machine learning methods, such as the random forest, can currently provide clearer answers regarding the importance of the features that they use and divulge thresholds by which the models make their best-performing classification methods, forming the basis of direct translation to clinical testing. Machine learning and its application to metabolomics and multi-omics data are reviewed in detail elsewhere [65,66].…”

Microbial Metabolites in Colorectal Cancer: Basic and Clinical Implications

“…In recent years, advanced analytic platforms including desorption electrospray ionization mass spectrometry (DESI-MS) [ 31 , 32 ], matrix-assisted laser desorption ionization imaging mass spectrometry (MALDI-MS) [ 33 ] and nanoscale secondary ion mass spectrometry (NanoSI-MS) [ 34 ] have been developed to comprehensively enhance the resolution and metabolites coverage of conventional MS-based method. In addition, with the development of machine learning and artificial interagency, advances in computational methods have greatly assisted metabolomics data processing, metabolite identification, as well as in metabolic phenotyping and biomarker discovery [ 35 , 36 ].…”

Coupling Machine Learning and Lipidomics as a Tool to Investigate Metabolic Dysfunction-Associated Fatty Liver Disease. A General Overview

“…In the age of “big” data, ML is a discipline in computer science, wherein machines (computers) can learn patterns from data, and the learned model(s) can be used to predict outputs [ 128 , 129 , 130 ]. In science and biomedicine, ML can find predictive patterns to understand complex biological systems and is currently used in lipidomics to process the amount of data generated by modern mass spectrometry [ 131 ]. In the context of metabolic studies, we can create a predictive model that predicts a given metabolite according to the peak detection and may improve diagnostic accuracy and treatment variability to make progress under a clinical approach [ 132 , 133 , 134 , 135 ].…”

“…DL techniques transform the data by iteratively tuning their internal parameters and may enable the extraction of the most predictive features from complex datasets. A selection of open-source tools for ML based on DL architectures may be found elsewhere [ 131 ].…”