Mapping Chemometrics with Chemometrics

During their daily practice of the topic, chemometricians often face the question “what is Chemometrics?”, asked by students, laboratory analysts, chemists, and industry customers. Much effort is spent trying to come up with clear and concise answers, also considering the different levels of detail at which the question needs to be answered. Due to the strong link between Chemometrics and experimental real-life problems, the application of statistical and machine learning methods is done from a practical perspective, which often goes well beyond the simple “trust” that a certain method will work. The “one-fits-for-all” approach is however very common in the wide field of data science, which also is, like all other fields of human knowledge and activities, subject to fashion trends. This often leads to posts on social networks (LinkedIn above all) of extremely simplified data analysis workflows, mainly done with the purpose of getting clicks and views, while providing a partial point of view, strongly influenced by the trendiest methods. Therefore, together with the question “what is Chemometrics”, we should probably ask ourselves: “where is Chemometrics?”. Starting from the considerations proposed by José Amigo in 2021 [1], we believe that a discussion about the relationship between Chemometrics and the wide field of data science (does the latter includes the former, or vice versa?) should be started in our community. This leads to a second important layer of the subject, which is about how to “organize” and compare the methods and topics of Chemometrics, according to their purpose(s) and uses, but also their characteristics. This would mean describing Chemometrics with the tools of Chemometrics. The discussion we would like to stimulate aims at getting a better understanding of our beloved field of research, starting from collecting the opinions of the participants to the CAC 2022 Conference. We propose to use a data collection method from the food sensory field, the Napping technique proposed in 2005 by Jérǒme Pagès [2]. This method allows non-expert panellists to evaluate products based on their own personal criteria: the products get placed on a tablecloth according to their perceived similarity (close) or dissimilarity (distant). The result is a perceptual map which can be digitalized and converted into a set of Euclidean distances. All data collected from the participants can be organized as a matrix which can be analyzed by Procrustes Multiple Factor Analysis [3], which provides, among the other outputs, a “consensus configuration”, which looks like a map with all evaluated objects on it.