From biological neurons to artificial neurons, the fascinating world of neural networks

Currently, the use of artificial intelligence (AI) has surpassed the limits of science fiction. It is an undeniable fact that it has an increasing boom in the development of sciences and technologies, as well as industrial processes, due to the fact that it increases efficiency, allows automation, among many other advantages.

In the scientific field, a branch of AI called Artificial Neural Networks (ANNs) stands out, since they are powerful tools for the analysis of complex signals and the modeling of systems in different areas such as neurosciences, mathematics, statistics, physics, computer science, engineering and recently chemistry. But what are ANNs?

There are different definitions, however, all of them agree that they are computational systems that are made up of a set of units called artificial neurons connected to each other and arranged in layers that are used to process and transmit information. These systems try to mimic the functioning of neurons in the brain, therefore, their structure is similar to that of a biological neural network, as shown in Figure 1.

Figure 1. Representation of a biological and an artificial neural network.

The learning of biological neurons in the brain is based on the acquisition of new knowledge through one's own or other people's experiences and the extraction of generic knowledge from a set of data. In contrast, artificial neurons, in order to learn and transmit a response, require a set of data and need to be programmed, since there are no stimuli from the environment that allow them to learn.

An ANN is organized in three main layers, the input layer that receives the data, the hidden layer (which can be several) that processes that data and the output layer from which the response of the network will be obtained. Through advanced mathematics, the ANN is able to learn the complex relationships in the data set it is fed with and even to recognize patterns. It is then tested with standard data to see if it is capable of reproducing them, if its responses vary greatly from what is expected and it restarts the process by optimizing it so that, little by little, it finishes learning until it achieves the expected result, this is what is known as ANN training.

ANNs have proven to be very useful, either alone or in combination with other methods, for solving complex classification, identification, diagnostic and quantification tasks. In analytical chemistry, ANNs are being used in conjunction with electrochemical techniques for the development of so-called electronic languages.

In these systems, by means of various sensors, a set of complex signals that can be interpreted by an ANN are recorded as shown in Figure 2, and can be applied in the analysis of the quality of wines and liquors, the detection of patterns in the agri-food sector, in systems for the detection of pollutants in the atmosphere or bodies of water, in the simultaneous and real-time analysis of different metabolites in blood for clinical diagnoses, in the control of chronic diseases, in the quality control of drugs, among others.

Schematic view of the operating mechanism of the electronic tongue. Reprinted from "A review on conjugated polymer-based electronic tongues," by P. Vahdatiyekta, M. Zniber, J. Bobacka, and T. P. Huynh, 2022, Analytica Chimica Acta, 1221, 340114.

The Materials Area of the Universidad Autónoma Metropolitana (UAM) Azcapotzalco unit, together with the Academic Area of Chemistry of the Universidad Autónoma del Estado de Hidalgo (UAEH), are developing and optimizing electrochemical methodologies based on miniaturized chemical sensors, developed with novel materials, which coupled to an ANN can perform the quantification of chemical compounds present in very complex samples, which under normal conditions would take a long time to perform at very high analysis costs. An example is the analysis of diclofenac sodium in commercial drugs where it can be accurately and precisely determined, even in the presence of other active substances such as paracetamol, pyridoxine and caffeine.

The use of RNAs still has a very wide field of application. With the development of increasingly modern software that allows the application of programming and training on personal computers, ANNs are gradually finding many niches of opportunity that allow support to solve health, environmental and food problems and thus improve the welfare of society in general.