AI’s Significance to the Future of Analytical Chemistry

Historically, the development of analytical chemistry has been a cornerstone of scientific analysis. From rudimentary titration apparatus and the Duboscq colorimeter (used for visual colour comparison) invented in the 19th century [2], to modern instrumentation like Acoustic Ejection Mass Spectrometry (AEMS) [3] and scanning ion conductance microscopy (SICM) that were found less than a decade ago, chemists have built increasingly powerful tools to measure the chemical composition of various substances. Yet in recent decades, a new bottleneck has emerged. Presently, the issue at hand now seems to not be collecting said information, but interpreting it. And thus, with the advent of artificial intelligence (AI), scientists hope to solve the issue of information interpretation with tools such as convolutional neural networks, random forests and language learning models [4].

What is Analytical Chemistry?

At its core, analytical chemistry is defined by the American Chemical Society as “the science of obtaining, processing, and communicating information about the composition and structure of matter [7].”  Fundamentally, analytical chemistry allows scientists to precisely understand the (almost) complete makeup of any substance, down to the number and type of atoms present as well as its structure. For example, a blood sample, pharmaceutical tablet and even microplastic traces in river water can all be identified and broken down into different chemical signals using techniques such as mass spectrometry, chromatography or photospectroscopy [8]. These techniques have been proven to be extremely accurate, capable of detecting trace amounts of specific molecules among thousands of others [9].

However, this incredible level of sensitivity also has many caveats. Specifically, the resulting datasets tend to be extremely large and noisy, with a single mass spectrum containing tens of thousands of peaks [11]. As such, this results in data interpretation being unviable at first glance. Consequently, chemists still need to heavily filter through the data in order to obtain the desirable molecules of interest. Traditionally, chemists highly rely on known reference signals and simplified statistical models to obtain useful insights from the results [12]. Yet, as analytical techniques become more complex and sophisticated, the ability to interpret results also becomes more challenging.

Example of data presentation in chemometrics, the data-driven study of chemistry [13].

Thus, AI excels precisely in recognising these complex patterns. Specifically, Machine Learning algorithms are trained on known spectra and thus learn to associate particular signal patterns with certain chemical molecules [14]. Crucially, this article wishes to delve into the 2 most common techniques applied by AI: random forests and convolutional neural networks.

The Basics of Machine Learning

At a simple level, random forests is a form of machine learning that combines many decision trees to make more reliable predictions, hence “combining the output of multiple decision trees to reach a single result” [15]. Each tree is trained on a slightly different version of the dataset, through randomly resampling the original data. Additionally, some data points can also be left out; otherwise known as “out-of-bag” samples, they can test how well the model performs. As such, this form of machine learning is incredibly useful when applied to results that are multi-variable and exist in large quantities, which humans have to otherwise take weeks or months to accurately analyse.

Next, for even more complex datasets, convolutional neural networks (CNNs) are utilised. A type of deep learning, CNNs treat spectra as visual patterns instead of numbers, identifying small recurring features in the input space [17]. Therefore, when applied to datasets, CNNs effectively treat multi-variable spectra as one-dimensional images. As a result, in the context of analytical chemistry, CNNs would hence be able to detect local patterns such as peak shapes or signal edges in mass spectrometry or photospectrometry.

Applications to Analytical Chemistry

Now that we understand how the mechanisms of machine learning algorithms work to simplify complicated information obtained from molecular analysis, we can explore some real-world case studies. In this article, we shall focus on the following 2 interesting uses of AI in aiding analysis: treating Alzheimer’s Disease and developing metal-organic frameworks.

Case Study 1: Alzheimer’s Disease

One of the most compelling reasons behind AI’s application in this field lies in accelerating medical diagnostics. In particular, analytical chemistry is used to discover biomarkers in patients with Alzheimer’s Disease (AD). A recent study applied a hybrid AI strategy, utilising LLMs to interpret neuroimaging data, genetic profiles and clinical information ofpatients with varying degrees of symptoms associated with Alzheimer’s disease [19]. Rather than depending on a single molecular marker, the model applied a multi-input convolutional neural network (MI-CNN) strategy to identify patterns within the aforementioned three datasets. Thus, this allows for the detection of minute genomic and phenotypic patterns linked to early AD symptoms, which were never discovered previously. From this study, the AI model (GenBERT) had an accuracy of 98.30%, far surpassing human capability in the early diagnosis of the neurodegenerative disease.

Case Study 2: Metal-organic frameworks

Another example of the many applications of AI in analytical chemistry may come as a surprise to readers: metal-organic frameworks (MOFs). Famous for helping scientists win the 2025 Nobel Prize in Chemistry, this material also has high potential for artificial intelligence integration. MOFs are defined as “hybrid inorganic–organic microporous crystalline materials formed from metal ions and organic linkers through coordination bonds” [21]. This means these structures allow for adjustable pore sizes and high surface area, providing uses in gas storage, catalysis and sensing.

Traditionally, researching new MOFs with desired chemical properties and accurately identifying their functions was a trial-and-error process, proving to be both tedious and at times unsuccessful. However, in 2024, a Nature study showcased a new AI system called ChatMOF, which predicted MOF properties and even produced new structures simply through natural language queries [23]. Through integrating various language models (GPT-4, GPT-3.5-turbo, llama2-7B-chat, llama-2-13B-chat), as well as MOF databases and MOF construction simulators, ChatMOF can combine functions of the various tools at hand to effectively respond to user queries about MOF characteristics or even propose candidate structures based on the MOF traits requested. Interestingly, ChatMOF achieves high accuracy in searching (96.9%), predicting (95.7%) and generating (87.5%) MOF frameworks, far surpassing traditional human experimentation and testing. Hence, this illustrates how AI can even serve as a catalyst for the discovery of relatively new structures like MOFs, speeding up the development of new technology through the applications of analytical chemistry.

Drawbacks

However, while AI does shine a glimmer of hope in the future of analytical chemistry, its relatively new implementation in performing analytical tasks also leaves plenty of limitations. For example, most analytical chemistry papers who successfully utilised AI to identify patterns, invent new chemical structures and so on are still unable to truly figure out the main reasoning behind why the AI chose such prediction. Commonly described to behave like a “black box”, chemists thus receive little meaningful insight into the responses from AI, making it difficult to obtain scientific understanding and certainty [25].

As of now, AI models also tend to over-rely on large, well-annotated datasets, which are not always realistic and unbiased in the context of analysing new, complex results from recent analysis [26]. Hence, AI models may also result in producing illegitimate patterns rather than true chemical relationships, ironically leading to research becoming misleading and even more difficult to verify.

Conclusion

To conclude, the role of AI in the field of analytical chemistry is more pertinent now than ever. From analysing biomarkers in patients with AD, to analysing and generating completely new MOF designs, to utilising machine learning from random forests to CNNs, AI acts as a new interpretive lens serving to complement established datasets made from real-world chemical analysis. Rather than replacing chemical intuition, it enhances it, allowing patterns and discoveries once unbeknownst to become unveiled to scientists (though with certain limitations). Thus, as AI’s role in analytical chemistry expands from here, we start to ponder upon how this new way of seeing chemistry will reshape what we choose to measure, understand, and ultimately define as chemical knowledge itself.

References:

[1]

C. Bradley, “Integrating AI and Machine Learning in Analytical Chemistry,” Lab Manager, Aug. 18, 2025. https://www.labmanager.com/integrating-ai-and-machine-learning-in-analytical-chemistry-34221

[2]

“Duboscq Colorimeter,” Si.edu, 2025. https://americanhistory.si.edu/collections/object/nmah_2502

[3]

H. Zhang et al., “Acoustic Ejection Mass Spectrometry for High-Throughput Analysis,” Analytical Chemistry, vol. 93, no. 31, pp. 10850–10861, Jul. 2021, doi: https://doi.org/10.1021/acs.analchem.1c01137.

[4]

C. Bradley, “The Future of Analytical Chemistry: Innovations on the Horizon,” Lab Manager, Aug. 29, 2025. https://www.labmanager.com/the-future-of-analytical-chemistry-innovations-on-the-horizon-34265

‌[5]

SciHistoryOrg, “Bausch & Lomb Duboscq Type Colorimeter,” Science History Institute Digital Collections, 2017. https://digital.sciencehistory.org/works/76537169x (accessed Jan. 13, 2026).

[6]

Danaher Life Sciences, “The Acoustic Ejection Mass Spectrometry System , SCIEX Echo®,” Bioprocessonline.com, 2026. https://www.bioprocessonline.com/doc/the-acoustic-ejection-mass-spectrometry-system-sciex-echo-0001

[7]

American Chemical Society, “Analytical Chemistry,” American Chemical Society. https://www.acs.org/careers/chemical-sciences/areas/analytical-chemistry.html

[8]

C. Bradley, “Analytical Chemistry Applications Across Industries,” Lab Manager, Aug. 27, 2025. https://www.labmanager.com/analytical-chemistry-applications-across-industries-34254

[9]

P. M. Nowak, R. Wietecha-Posłuszny, and J. Pawliszyn, “White Analytical Chemistry: An approach to reconcile the principles of Green Analytical Chemistry and functionality,” TrAC Trends in Analytical Chemistry, vol. 138, p. 116223, May 2021, doi: https://doi.org/10.1016/j.trac.2021.116223.

[10]

Broad Institute, “What is Mass Spectrometry?,” Broad Institute, Sep. 13, 2010. https://www.broadinstitute.org/technology-areas/what-mass-spectrometry

[11]

G. M. Merone et al., “Analytical Chemistry in the 21st Century: Challenges, Solutions, and Future Perspectives of Complex Matrices Quantitative Analyses in Biological/Clinical Field,” Analytica, vol. 1, no. 1, pp. 44–59, Oct. 2020, doi: https://doi.org/10.3390/analytica1010006.

[11]

I. G. Zenkevich, D. A. Nikitina, and A. S. Kushakova, “Processing and Interpretation of Analytical Data with a High Degree of Uncertainty,” Springer Nature Link, vol. 77, no. 11, pp. 1399–1412, Nov. 2022, doi: https://doi.org/10.1134/s1061934822090143.

[12]

Rafael Cardoso Rial, “AI in analytical chemistry: Advancements, challenges, and future directions,” Talanta, vol. 274, no. 125949, pp. 125949–125949, Jul. 2024, doi: https://doi.org/10.1016/j.talanta.2024.125949.

[13]

Archita Patel, “Ultraviolet spectrophotometry (dual wavelength and chemometric) and high performance liquid chromatography for simultaneous estimation of meropenem and sulbactam sodium in pharmaceutical dosage form,” researchgate.net, 2014. https://www.researchgate.net/figure/Calibration-and-validation-concentration-data-set-for-chemometrics_tbl1_260029942 (accessed Jan. 13, 2026).

[14]

Rafael Cardoso Rial, “AI in analytical chemistry: Advancements, challenges, and future directions,” Talanta, vol. 274, no. 125949, pp. 125949–125949, Jul. 2024, doi: https://doi.org/10.1016/j.talanta.2024.125949.‌

[15]

IBM, “What is random forest?,” Ibm.com, Oct. 20, 2021. https://www.ibm.com/think/topics/random-forest

[16]

A. Jain, “Everything about Random Forest - Abhishek Jain - Medium,” Medium, Sep. 14, 2024. https://medium.com/@abhishekjainindore24/everything-about-random-forest-90c106d63989

[17]

P. Ratan, “Convolutional Neural Network Made Easy for Data Scientists,” Analytics Vidhya, Oct. 28, 2020. https://www.analyticsvidhya.com/blog/2020/10/what-is-the-convolutional-neural-network-architecture/

[18]

IBM, “Convolutional Neural Networks,” Ibm.com, Oct. 06, 2021. https://www.ibm.com/think/topics/convolutional-neural-networks

[19]

J. Song, J. Huang, and R. Liu, “Integrating NLP and LLMs to discover biomarkers and mechanisms in Alzheimer’s disease,” SLAS Technology, vol. 31, p. 100257, Apr. 2025, doi: https://doi.org/10.1016/j.slast.2025.100257. priority_highIssue number

close

[20]

“Our Science | Acumen Pharmaceuticals,” Acumen, Mar. 30, 2021. https://acumenpharm.com/science/ priority_highWebpage author

close

[21]

“Metal Organic Framework - an overview | ScienceDirect Topics,” Sciencedirect.com, 2018. https://www.sciencedirect.com/topics/chemistry/metal-organic-frameworkpriority_highWebpage author

close

[22]

J. Ren, N. M. Musyoka, H. W. Langmi, A. Swartbooi, B. C. North, and M. Mathe, “A more efficient way to shape metal-organic framework (MOF) powder materials for hydrogen storage applications,” International Journal of Hydrogen Energy, vol. 40, no. 13, pp. 4617–4622, Apr. 2015, doi: https://doi.org/10.1016/j.ijhydene.2015.02.011.

[23]

Y. Kang and J. Kim, “ChatMOF: an artificial intelligence system for predicting and generating metal-organic frameworks using large language models,” Nature communications, vol. 15, no. 1, Jun. 2024, doi: https://doi.org/10.1038/s41467-024-48998-4.

[24]

A. Ozcan et al., “Artificial Intelligence Paradigms for Next-Generation Metal–Organic Framework Research,” Journal of the American Chemical Society, vol. 147, no. 27, pp. 23367–23380, Jun. 2025, doi: https://doi.org/10.1021/jacs.5c08214.

[25]

Adrián Fuente-Ballesteros, V. Samanidou, Seyed Mosayeb Daryanavard, A. M. Ares, and J. Bernal, “Artificial Intelligence as a Scientific Copilot in Analytical Chemistry: Transforming How We Write, Review, and Publish,” Analytical Chemistry, Sep. 2025, doi: https://doi.org/10.1021/acs.analchem.5c03767.

[26]

Tobias Elsbecker, “Limitations of databases and opportunities for Artificial Intelligence - 2025 - Wiley Analytical Science,” Analytical Science Article DO Series, Apr. 2025, doi: https://doi.org/10.1002/was.000600952.