Chemometrics reveals not-so-obvious analytical information

Abstract

The application of chemometric tools in analytical chemistry or other areas of chemistry has become essential. This is mainly due to the large amount and nature of the generated data1,2 and the need to extract useful information from these and optimize steps throughout a process. It allows the quick decision-making visualization of interactions among variables, such as synergism or antagonism between parameters, during the development of a method,3 as shown in Figure 1. Classical chemometric techniques have been disseminated and can be divided according to the study approach, among which exploratory data analysis stands out. Principal component analysis (PCA) is one of the most accessible and well-established ways to perform an initial exploration and extract relevant information from a given dataset and has been used quite successfully in various spectroscopic techniques.4 Principal component analysis consists of projecting the data in a smaller dimension, enabling the detection of anomalous samples (outliers), the selection of essential variables in a given system, and unsupervised classification.1,2,4 Another branch of chemometrics involves the design of experiments (DoE). The primary purpose of the factorial design is to study the influence or effect of a given variable and its interactions in a specific system.5-9 Multivariate calibration is another aspect of chemometrics, where several variables are used to calibrate one (or more) property or the concentration of a given chemical analyte.10,11 Since the first publications of chemometric tools, numerous variations of these techniques, proposals for data fusion strategies, and applications using hyphenated instrumental techniques have been proposed.12-14 Industrial quality control and development (R&D) laboratories require an approach addressing adequate quality by design (QbD). The QbD strategies consider four steps that include an analytical target profile (ATP), a risk assessment, a design space (DS), and control strategy and validation based on figures of merit, for instance.9 Principal component analysis is the most widely multivariate technique used for data analysis. Jolliffe wrote a review reporting his wonderful experience with PCA in the last 50 years.15 Indeed, PCA is an invaluable method for data, and I agree with it. PCA is the algorithm of choice for numerous chemometric techniques.16 Other computational languages, such as Python, are currently experiencing a rise in popularity in the field of chemistry. The R language has also become more popular than it was ten years ago. The scripts, functions, or codes are easily written with fewer lines and specific commands that minimize steps and help speed up calculations. The dissemination of free software has also become popular, and the sharing of codes through publications, social media, communities, or websites has become relatively easy. From my point of view, chemometrics is no longer faced as a giant monster or a way to become scientific papers fancier without helpful content. Chemometrics extract information that is not easy to visualize at first through univariate evaluations or using simple plots. Nowadays, thousands of instrumental data can furnish important chemical information, and we must use them for significant proposals. Indeed, QbD is proof of that, as Industry 4.0 is a reality.