Determination of Libidibia ferrea Markers Using Spectrophotometry and Chemometric Tools with Comparison to a Standard High-Performance Liquid Chromatography (HPLC) Protocol

Abstract

This study evaluates the performance of different regression algorithms based on artificial intelligence in the quantification of ellagic and gallic acids from ultraviolet (UV) data. Samples of Libidibia ferrea, which is known for its medicinal properties and mostly consumed as tea, were used. The medicinal properties are related to gallic and ellagic acids present in the species. Thus, quantifying gallic and ellagic acids is important to differentiate the biological activities. This quantification is generally performed by high-performance liquid chromatography (HPLC). However, HPLC is expensive and uses polluting solvents. A potential alternative to HPLC is spectrophotometry, which is less expensive and easier to implement. However, the spectrophotometric signal has many overlapping components. This characteristic makes it difficult to quantify substances by spectrophotometry. Computational tools and chemometric methods are used to overcome this limitation. To perform the analysis, eight regression models were used: linear regression, support vector machine, random forest, random tree, and four configurations of extreme learning machines, with kernel: linear, third-degree polynomial, sigmoid, and sine. The Pearson correlation coefficient () was the main metric used to evaluate the performances of the algorithms. The regression algorithms were shown to deliver effective concentration predictions even from raw spectrophotometric measurements, providing results closer to the values using HPLC, with a maximum r value of 0.93 for ellagic acid and 0.83 for gallic acid. The overall findings are promising, pointing to the development of a low-cost and efficient quality control technique.