Key Points:
- Artificial intelligence and machine learning are revolutionizing chromatography by enabling more efficient method development, real-time instrument control, and deeper data interpretation.
- However, rather than replacing scientists, AI should be seen as an augmentation tool—enhancing human decision-making while preserving the theoretical and mechanistic understanding that underpins separation science.
- Before AI can yield reliable insights, chromatographic data must be harmonized and contextualized across different instruments and software platforms.
Chromatography, a cornerstone of analytical chemistry, has long relied on empirical methods and expert intuition. However, the advent of artificial intelligence (AI) is ushering in a new era of precision, efficiency, and discovery (1). It is an exciting era to be in as a separation scientist, but with this excitement comes a trepidation of what the future will bring—whether there will still be a need for human expertise, or if advances in automation will fully take over tasks such as method development and data interpretation.
The Evolution of Analytical Paradigms
Artificial intelligence and machine learning (ML) fundamentally reshape scientific methodology from a deductive to an abductive paradigm (2). Traditional practices have followed a top-down approach: researchers formulate theories (like the van Deemter equation) that explain chromatographic behavior through mathematical relationships between variables such as flow rate, particle size, and temperature (3). This deductive method begins with theoretical principles that are subsequently tested with targeted experiments. By contrast, AI/ML transforms science into a sophisticated comparative problem—algorithms explore vast multidimensional data spaces to discover patterns and relationships that human intuition might never conceive (4).
AI-based science represents both a technical advance and an epistemological shift from traditional science, but the two approaches are complementary. ML-discovered patterns often inspire new theoretical frameworks, while established theory guides the meaningful interpretation of computational discoveries.
Chromatography in all of its forms—gas, supercritical fluid, liquid, and even electrophoretic—provides an excellent opportunity to perform this abductive analysis because it is a source of a vast amount of data, from environmental to pharmaceutical. The introduction of spectroscopic detectors such as mass spectrometry (MS) as a detection system for chromatography systems has further enhanced the amount of generated data, particularly high-resolution mass spectrometry (HRMS), which can produce enormous volumes, typically gigabytes or tens of gigabytes, of data in a single run.
The Data Harmonization Challenge
One of the significant challenges chromatographers face is to harmonize and contextualize the data so it can be entered effectively into the statistical models that help them make sensible deductions. Harmonizing ensures the data are available in the same format, and contextualizing ensures there’s metadata attached (metadata is the information associated with the sample or chromatographic operating conditions, but does not correspond to the detector response). All manufacturers of analytical instrumentation have developed proprietary software to operate the instruments, and they have also introduced unique data formats that do not lend themselves to integrating with other software, particularly other chromatography data systems (CDS) software, and certainly not with statistical packages.
Attempts to standardize CDS data formats have led to initiatives like the Allotrope Foundation (5). The Allotrope Simple Model (ASM) is a standardized, lightweight, and human-readable way to structure instrument data and simplify data exchange and laboratory analysis.Employing a tabular format with key‑value pairs derived from the Allotrope Foundation Ontology (AFO), ASM promotes consistency and facilitates the aggregation and indexing of related data (6). However, despite its advantages, the ASM faces challenges such as the need for broad industry adoption, the complexity of mapping diverse instrument outputs to a common standard, and the ongoing effort required to expand its coverage across various analytical techniques while ensuring semantic accuracy and community consensus.
Standardization of the data allows for the data to be easily reviewed and, if necessary, normalized, with obvious outliers being removed from data sets if necessary. Applying machine learning without this critical step will produce very dubious results. The adage of “garbage‑in‑garbage-out” applies very strongly here, as even though it is typically very large data sets being interpreted, outliers can significantly impact a statistical model’s conclusions. However, once this critical step has been performed, AI can be applied to an array of applications in the world of separation science.
Understanding AI Approaches in Separation Science
AI encompasses a spectrum of computational approaches—from basic regression to advanced neural networks and deep learning. These methods fall into three fundamental categories with specific applications in chromatography.
1. Supervised learning uses labeled data to establish relationships between inputs and outputs, making it ideal for method optimization when target outcomes are known. For example, predicting retention times based on mobile phase composition or optimizing resolution based on temperature and pH adjustments (7).
2. Unsupervised learning identifies intrinsic patterns within unlabeled data sets, revealing hidden structures without predefined classifications. This approach excels at finding patterns in complex chromatograms from metabolomics or proteomics, grouping similar compounds, or detecting anomalies in large data sets (8).
3. Reinforcement learning iteratively improves models through feedback loops, where algorithms learn optimal behaviors by maximizing rewards based on their actions. In chromatography, this can be applied to adaptive method development, where separation parameters are continuously adjusted based on peak quality metrics (9).
While choosing the appropriate model may initially seem daunting, remember that all AI approaches share one fundamental purpose: to uncover meaningful relationships within your separation data that enable, for example, accurate predictions and process optimizations, whether through simple regressions or complex multi-layered neural networks.
Chromatographic data, especially from complex samples, piles up in overwhelming amounts. But no matter the size of the data set, AI excels at pattern recognition and multivariate analysis, enabling the identification of subtle peaks, quantifying analytes, and deconvoluting overlapping signals. Researchers in the omics field, for example, commonly use an ML method called principal component analysis, or PCA, to find patterns in data in very complex data sets (8). Chromatographers are already using pattern recognition and predictive models based on large, complex data sets, but perhaps we are not so aware that we are straying into the field of AI.
Current Applications of AI in Chromatography
AI is already making significant strides in various aspects of chromatography, addressing persistent challenges and enhancing analytical workflows in the following areas:
Method Development and Optimization: Traditional method development is often time-consuming and resource-intensive (9). AI algorithms, particularly neural networks and support vector machines, can analyze large data sets of chromatographic parameters (for example, mobile phase composition, temperature, flow rate) and predict optimal separation conditions. The extensive online literature on this kind of work makes applying the AI models significantly easier and could dramatically reduce the amount of effort required to develop a method.
AI can also reduce solvent consumption and improve the robustness of separations. However, its use does require a high level of control over variables such as column consistency. AI will also highlight when methods are intrinsically unstable due to the experimental parameters being too close to a cusp, such as a pKa.
AI-powered software can perform simulations to predict the outcome of different parameters, thus reducing the number of real-world experiments needed (10). It should be stated that there are theoretical models, not based on AI, that have been developed into commercial software packages that will allow for accurate retention time prediction based on just a few experiments.
Instrument Control and Automation: AI-powered systems can monitor instrument performance, detect anomalies, and adjust parameters in real time to maintain optimal separation conditions. This is invaluable in a quality control (QC) environment where changing a column can be impactful due to equilibration issues, as it will allow for optimizing column and instrumentation use. This results in improved instrument uptime, reduced maintenance and consumable costs, and increased data reliability. AI can also be used for predictive maintenance, anticipating when parts of the machine may need to be replaced and thus avoiding downtime and unnecessary costs.
Compound Identification and Library Searching: AI algorithms can analyze spectral data (for example, mass spectra, UV–vis spectra) obtained from chromatographic detectors and compare them to spectral libraries for rapid compound identification. This accelerates the identification process, particularly for unknown compounds. AI can improve the accuracy of library searching by taking into account the context of the chromatographic separation.
Future Applications of AI in Chromatography
The future of chromatography promises even more transformative applications of AI, leading to breakthroughs in various fields.
Personalized Chromatography: AI could enable the development of personalized chromatographic methods tailored to specific sample matrices and analytical goals. This would be particularly beneficial in fields such as personalized medicine, where individual patient samples require customized analysis.
Real-time Process Monitoring and Control: AI-powered systems could enable real-time monitoring and control of industrial chromatographic processes, ensuring consistent product quality and optimizing production efficiency. This will have a significant impact on pharmaceutical production (11).
Hyperspectral Imaging Integration: Combining hyperspectral imaging with AI, alongside chromatographic techniques, will allow for very detailed analysis of complex samples. AI will be needed to process the large amount of data generated (12).
Discovery of Novel Separations: AI algorithms could explore vast chemical spaces to discover novel stationary phases and mobile phase combinations, leading to the development of more efficient and selective, even greener, separations (13). This has the potential to find optimal solutions to difficult separations.
Autonomous Chromatography: By developing fully autonomous chromatographic systems that perform method development, sample analysis, and data interpretation with minimal human intervention, we can free up valuable time for researchers and analysts. Scientists could focus on the kind of work they enjoy, such as experimental design, novel applications, interpretation of the significance of their data, or developing new theoretical frameworks.
Integration With Other Analytical Techniques: From the onset of the development of chromatography, it has been essential to couple the technology with other analytical technologies to allow detection of the separated components. This has seen a journey from simple detection systems such as colorimetry and thermal conductivity to HRMS and even nuclear magnetic resonance (NMR) (14). AI makes combining chromatography with other analytical techniques, such as MS and NMR, easier to interpret the data. This will allow for a more complete understanding of samples and begin to aid predictive models.
Limitations and Challenges
Despite its transformative potential, the application of AI in chromatography faces several challenges related to model interpretability and transparency. Many advanced AI models operate as “black boxes,” making it difficult to understand the reasoning behind their predictions. This lack of transparency can be problematic in regulatory environments where method validation requires a clear understanding of operating principles and the variables’ impact on the chromatographic output. Scientists must weigh the benefits of improved performance against the loss of mechanistic clarity that has traditionally characterized chromatographic method development. This is particularly important in regulated industries like pharmaceuticals, where authorities may require comprehensive documentation of the methodology (15).
The scale and complexity of model development present significant technical hurdles for chromatography laboratories. Modern deep learning approaches often require specialized expertise in data science and substantial computational resources that may be beyond the reach of many analytical laboratories. In addition, the cross-system validation of AI models remains problematic—algorithms trained under idealized conditions may struggle with the variability encountered in real-world samples, and models developed on one instrument platform may perform poorly when transferred to different hardware. These practical limitations can significantly slow adoption in routine analytical environments where robustness and reliability are paramount, particularly in QC settings where method transfer between sites is common.
Perhaps the most profound challenge is balancing technological advancement with fundamental scientific understanding. An overreliance on AI-driven approaches could potentially erode the deep theoretical understanding that has driven advances in separation science for decades. When algorithms replace understanding, there’s a risk that future generations of separation scientists may lose touch with the physicochemical principles that allow them to understand specific experimental scenarios. This could ultimately limit innovation, as breakthrough developments often arise from deep theoretical insights rather than purely empirical approaches. Maintaining this balance between leveraging AI’s pattern-finding power while preserving and advancing fundamental understanding will be essential for the healthy evolution of separation science.
The Augmented Chromatographer
AI is poised to transform chromatography from an empirical science to a data-driven and automated discipline, but this transformation should be viewed as augmentation rather than replacement. The future separation scientist will likely become more of an interpreter and strategist, using AI tools to push the boundaries of what’s possible while applying the fundamental principles that have guided the field for generations.
By addressing the challenges and embracing the opportunities, we can unlock the full potential of AI to drive innovation and discovery in separation science and beyond. For the separation scientist, this will mean learning new skills, focusing on the implications of the data interpretation and less on the routine sample and data analysis. As separation scientists, we must embrace this future to drive efficiency and make our science more robust. The greater use of separation science will also further highlight the importance of our science and the benefits that chromatographers have provided to society for more than a century. However, we must ensure we do not lose the hard-won, fundamental science developed over that period. And we must ensure we nurture the enthusiasm of the next generation of separation scientists to take on board the benefits of AI to improve our understanding of the world around us.
References
(1) Taylor, T. Generative AI in Chromatographic Analysis. CHEManager Intern. 2024, 24–25. https://chemanager-online.com/en/news/generative-ai-in-chromatographic-analysis#:~:text=19.03.2024%20%E2%80%A2%20News%20%E2%80%A2,Artificial%20Intelligence%20(AI)%20%E2%80%A2%20Chromatography (accessed 2025-05-09).
(2) Kingsley, O.-A. Artificial Intelligence Research: A Review on Dominant Themes, Methods, Frameworks and Future Research Directions. Telemat. Inform. 2024, 14, 00127. DOI: 10.1016/j.teler.2024.100127
(3) Van Deemter, J. J.; Zuiderweg, F. J.; Klinkenberg, A. Longitudinal Diffusion and Resistance to Mass Transfer as Causes of Non Ideality in Chromatography. Chem. Eng. Sci. 1956, 5 (5), 271–289.
(4) Theobald, O. Machine Learning For Absolute Beginners: A Plain English Introduction (Second Edition); 2017.
(5) The Allotrope Foundation. https://www.allotrope.org/ (accessed 2025-05-01).
(6) Allotrope Foundation Introduction to the Simple Model (ASM). https://www.allotrope.org/introduction-to-allotrope-simple-model (accessed 2025-05-01).
(7) Barron, L. P.; McEneff, G. L. Gradient Liquid Chromatographic Retention Time Prediction for Suspect Screening Applications: A Critical Assessment of a Generalised Artificial Neural Network-based Approach Across 10 Multi-residue Reversed-phase Analytical Methods. Talanta 2015, 147, 261–270. DOI: 10.1016/j.talanta.2015.09.065
(8) Theodoridis, G.; Gika, H.; Raftery, D.; et al. Ensuring Fact-Based Metabolite Identification in Liquid Chromatography–Mass Spectrometry-Based Metabolomics. Anal. Chem. 2023, 95, 3909–3916. DOI: 10.1021/acs.analchem.2c05192
(9) Niezen, L. E.; Libin, P. J. K.; Cabooter, D.; Desmet, G. Reinforcement Learning for Automated Method Development in Liquid Chromatography: Insights in the Reward Scheme and Experimental Budget Selection. J. Chrom. A 2025, 1748, 465845. DOI: 10.1016/j.chroma.2025.465845
(10) Xie, J.; Chen, S.; Zhao, L.; Dong, X. Application of Artificial Intelligence to Quantitative Structure–Retention Relationship Calculations in Chromatography. J. Pharm. Anal. 2025, 15 (1), 1–15. DOI: 10.1016/j.jpha.2024.101155
(11) Anupa, A.; Jesubalan, N. G.; Trivedi, R.; et al. Implementation of Machine Learning Tool for Continued Process Verification of Process Chromatography Unit Operation. J. Chromatogr. A 2025, 1742, 465642. DOI: 10.1016/j.chroma.2024.465642
(12) Yang, C.; Guo, Z.; Barbin, D. F.; et al. Hyperspectral Imaging and Deep Learning for Quality and Safety Inspection of Fruits and Vegetables: A Review. J. Agric. Food Chem. 2025, 73 (17), 10019–10035. DOI: 10.1021/acs.jafc.4c11492
(13) Dangayach, R.; Jeong, N.; Demirel, E.; et al. Machine Learning-Aided Inverse Design and Discovery of Novel Polymeric Materials for Membrane Separation. Environ. Sci. Technol. 2025, 59 (2), 993–1012. DOI: 10.1021/acs.est.4c08298
(14) Wilson, I. D.; Brinkman, U. A. Th. Hype and Hypernation: Multiple Hyphenation of Column Liquid Chromatography and Spectroscopy. TrAC 2007, 26 (9), 847–854. DOI:10.1016/j.trac.2007.07.007
(15) ICH, Q14 Analytical Procedure Development, (2023) https://database.ich.org/sites/default/files/ICH_Q14_Guideline_2023_1116_1.pdf
Tony Edge is a scientific business analyst working at TetraScience. He has worked in both manufacturing and also industry, having periods of employment at LGC and also AstraZeneca as well as ThermoFisher Scientific, Agilent Technologies, and latterly Avantor. In 2008, he was fortunate enough to be awarded the Desty memorial lecture for his contributions to innovating separation science, and in the same year also won a clinical excellence award from AstraZeneca. He was awarded an honorary fellowship at the University of Liverpool, where he lectured on separation science. He is also the president of The Chromatographic Society in the UK and a permanent member of the scientific committee for the International Symposium on Chromatography.
Daniela Pedersen is VP of Scientific Product Marketing at TetraScience Inc. She has held various positions in marketing, sales, and business development, including at Dassault Systèmes Biovia, Waters, and Bio-Rad Laboratories. Daniela has a long tenure of expertise in life sciences, laboratory, and scientific informatics, as well as in quality and regulatory compliance. Today she is responsible for the strategic messaging and positioning of TetraScience’s products and the development of marketing content with a strong focus on customer value and industry needs.
Bruce Upbin is VP, Communications at TetraScience Inc.
