The Use of Machine Learning in the Analytical Control of the Preparations of Medicinal Plants

Resumo

Despite the fact that the global market for medicinal plants amounts to hundreds of billions of dollars, there is almost no government control over the quality of such pharmaceuticals in most countries of the world. This is partly attributed to the complex composition of plant materials: traditional analytical methodology is based on the use of standard reference samples for each analyte. In this case, preparations based on medicinal plants may contain tens and hundreds of physiologically active components. Isolation of those compounds in a pure form in practice is carried out using preparative chromatography, which leads to their high cost. Moreover, variation of the chemical composition of medicinal plants depending on the geographical origin of the raw materials interferes with prescribing strict ranges of permissible contents for all physiologically active components. The combination of the above factors limits the possibilities of using traditional approaches to analysis requiring strict standardization, the list of compounds for each type of plant, levels of contents, and the availability of reference materials and standards of comparison. This led to the study of the possibility of introducing various mathematical approaches as an auxiliary methodology. Unlike traditional methodologies, machine learning approaches are based on the correct collection of the data samples. Such a sample should contain groups of samples that correspond to the states of the object which the developed algorithm must distinguish: authentic/fake, pure/containing impurities, effective/not containing a certain level of active components, etc. This review is devoted to consideration of the application of machine learning techniques to the problems of chemical analysis and production control of raw materials of medicinal plants and preparations on their basis for the last 15 years.