How AI tools are helping scientists read mass spectrometry data

The science

Original title

Artificial Intelligence and Machine Learning in Mass Spectrometry-Based Proteomics and Metabolomics.

Authors

Basse K, Szeitz B, Suits F et al.

Journal

Methods in molecular biology (Clifton, N.J.)

Published

Oct 15, 2026

Study type

Study · in humans

PMID

42878764 ↗

DOI

10.1007/978-1-0716-5519-1_13 ↗

Abstract

Analytical chemistry, and particularly mass spectrometry (MS), plays a central role in the molecular characterization of biological and environmental samples. These data are essential for understanding the molecular mechanisms underlying healthy aging, disease onset and progression, and the complex chemical composition of environmental systems. Over the past decades, MS has undergone unprecedented technological advancements. Current state-of-the-art instruments enable the  thorough characterization of deep proteomes at the single-cell level and achieve comprehensive coverage of metabolites and proteins. Innovations such as new mass analyzers (e.g., the Astral mass analyzer) and the integration of multidimensional separation techniques (e.g., trapped/cyclic ion mobility) have significantly expanded the analytical capabilities of the instrumentation. However, the volume and complexity of the data produced by these advanced platforms pose major challenges for interpretation and integration. To address these challenges, machine learning (ML) and artificial intelligence (AI) methods are emerging as indispensable tools in analytical chemistry. These approaches, particularly deep learning, generative AI, and large language models, offer powerful solutions for analyzing, integrating, and interpreting the high-dimensional, heterogeneous datasets generated in proteomics and metabolomics research. This chapter first introduces the fundamental principles and operational mechanisms of commonly used AI/ML methods in analytical chemistry. These include regression models, decision trees, k-nearest neighbors, support vector machines, neural networks, deep learning architectures, and generative AI. The first part of the chapter also discusses key requirements for model training, including the need for sufficient sample and molecular coverage, and evaluates how data noise affects model performance and reliability. In the second part, a series of representative applications are presented, demonstrating the use of AI/ML for peptide and metabolite observability prediction, retention time and ion mobility prediction, fragment ion spectrum generation, de novo compound identification, and improvements in compound quantification and scoring of identifications. The chapter concludes with practical recommendations for applying AI/ML approaches in proteomics and metabolomics studies, and for their integration into bioinformatics pipelines to enhance data interpretation, reproducibility, and scientific discovery.

Read it on PubMed ↗