recent articles analytical science (1) by Paolo Oliveri from the University of Genoa (Italy) discusses the growing influence of artificial intelligence (AI) in chemistry, arguing that tools now called AI and machine learning (ML) have been used in the chemometry field for many years. In his opinion, the real challenge is balancing the enthusiasm for new ML techniques with the chemometry tradition of understanding experimental systems, data quality, and preprocessing. Oliveri believes that by marrying rigorous chemical insights with new AI tools, the field of chemometry is poised for significant growth in modern analytical laboratories.
Oliveri sat with me. spectroscopy To dig deeper into the subject.
How do you think the early chemometric chemists mentioned in your article, like Svante Wold and Bruce Kowalski, view today's artificial intelligence (AI) and machine learning (ML) revolution in analytical chemistry?
Given their positive attitude towards the implementation of powerful computational methods and innovation in general, I believe that they will see the current AI revolution as a valuable opportunity to further promote the use of chemometrics in chemical laboratories. They will likely focus on improving and streamlining chemical processes, from method development to validation, monitoring, and quality control. And they will certainly acknowledge the central role of chemistry and chemists in this context. In an editorial article celebrating the first 20 years of chemometrics, Svante Wold writes: […] And chemometrics must continue to be motivated by solving chemical problems rather than developing methods. ” (2).
The key difference you cite is that chemometrics prioritizes. interpretabilityModern AI, on the other hand, often embraces “black box” models. How can chemical analysts leverage the predictive power of deep learning while maintaining chemical interpretability?
There is no doubt that the ability to interpret a model is important because it allows the experimenter to understand which experimental variables are responsible for a particular result, whether positive or negative. Interpretability is typically lacking in deep learning models, but is often fundamental to chemists. For example, when processing a set of spectra, you not only need to obtain qualitative or quantitative analysis results, but also understand which spectral features are associated with relevant information. This interpretation step is not only fundamental to the validation process of data-driven modeling and ensures that overfitting of the data is avoided, but is also an important part of the knowledge acquisition itself. Chemometricians who wish to benefit from the implementation of deep learning models must design modifications to such algorithms to achieve model interpretability by extracting the values of model coefficients and converting them into parameters that describe the contribution of each variable.
You emphasize that deep learning's demand for large datasets does not mesh well with typical chemical data collection. What strategies and innovations do you think will emerge to make AI methods more viable for small, high-quality chemical datasets?
Yes, another big problem with implementing deep learning techniques is the need for large and representative training datasets to create efficient models. One approach to overcome this hurdle is artificial data augmentation. This approach is interesting, but I have yet to see a completely satisfactory application. However, this is certainly a direction that could be further investigated and improved.
You mentioned that the number of chemometry courses is increasing at all academic levels. How should we train the next generation of chemometric chemists to survive in a world increasingly dominated by AI and data science?
It is necessary to distinguish between two levels of education. One for chemometrics users and one for developers. Users need to learn the key concepts needed to properly apply chemometrics, such as the importance of data quality and representativeness, how to avoid overfitting, how to properly validate models, and how to correctly interpret model results. Developers, on the other hand, need to dig deeper into statistical and ML theory, keeping in mind chemical constraints and the points mentioned above. However, just as you don't have to be an electronics engineer to use a smartphone, you don't have to be a chemical measurement developer to use chemical measurement techniques.
You emphasize the need for strong collaboration between chemometric chemists and data scientists in general. What would an ideal partnership between these two communities look like, and what challenges would need to be overcome to make it work?
This type of collaboration is beneficial to chemometers who can borrow advanced methods and adapt them to the nature of the chemical data and problem. Data scientists may then draw inspiration from real-world problems to develop targeted techniques and optimized algorithms that address specific needs. Problems related to slightly different technical languages used by different communities can be easily overcome by working together.
You describe data preprocessing as an important and underappreciated part of analysis. In the era of automated AI pipelines, how can we ensure that domain-specific knowledge, such as proper preprocessing, remains at the heart of data analysis?
Once again, we need to start by fully understanding the nature of chemical data and its properties. For this reason, the education of new generations of chemometric chemists must focus firmly on chemical aspects and their implications. There is a danger of paying too much attention to the mathematical and algorithmic aspects. The mathematical and algorithmic aspects are important, but they are secondary.
You quote Waldo's advice. “We need to remain chemists and adapt statistics to chemistry, not the other way around.” Given the dominance of computation and algorithms in current research, how important do you think that principle is still today?
This principle is essential for achieving profitable results from the implementation of machine learning in the chemical field. The previously proposed proposal to modify deep learning algorithms to extract variable importance parameters from model coefficients is a concrete example of this concept.
You contrast “mature” chemometric chemists who resist AI terminology with more enthusiastic younger researchers. How do you think these intergenerational conversations may shape the future direction of chemometrics?
The key to leveraging cutting-edge, state-of-the-art data processing tools, starting from a deep knowledge of the data characteristics, taking into account all practical constraints, and without losing focus on addressing real-world chemical problems, lies in combining the skills and concrete approaches of experienced professionals with the passionate impulses of the younger generation.
You conclude on an optimistic note about the convergence of AI and chemometrics. What specific developments or breakthroughs do you foresee that could define the “AI chemometrics” era over the next 10 years?
In fact, chemometrics is AI. We will definitely see integration with new AI frames such as generative tools, language models, multimodal approaches, and agent implementations. The next decade is expected to see significant increases in the automation and speed of data processing, and the widespread availability of sophisticated software tools. Therefore, it becomes even more important to invest time in a solid understanding of the experimental system and data characteristics to draw truly powerful applications.
References
- Oliveri, P. Chemometrics: A bridge to the AI era. analytical scientist 2025.
https://url.us.m.mimecastprotect.com/s/IzpjCJ6R2Ri0vB8ktkH5CyZoE5?domain=urlsand.esvalabs.com (Accessed 2025-11-06) - Wold, S. Chemometrics. What do we mean by it and what do we want from it? Kemama. intelligence. Laboratory systems. 1995, 30109-115. Doi:
/10.1016/0169-7439(95)00042-9
