A chemometrics-based approach for the determination of thymoquinone from Nigella sativa L. (Black Cumin) seeds of different geographical regions using the HPLC technique

Background and Aims: In this study, thymoquinone (TQ) from black cumin will be quantified from several geographical regions, including India, Syria, Saudi Arabia, Iraq, and Turkey. Additionally, to forecast the chromatographic behavior of the analyte in artificial intelligence (AI)-based models, the study used both ensemble machine learning methodologies and chemometrics-based approaches.Methods: An Agilent Technologies (1200 series, USA) instrument that includes an autosampler, a binary pump, a diode array detector (DAD), and a vacuum degasser was used for the HPLC analysis. Using five different single models-principal component regression (PCR), least square-support vector machine (LSSVM), least square boost (LSQ-BOOST), adaptive neuro-fuzzy inference system (ANFIS), and step-wise linear regression-the HPLC-DAD technique was used to simulate the qualitative and quantitative properties of TQ (SWLR). Results: The collected results demonstrated that samples from India and Iraq have the highest concentration of TQ. TQ was present in all samples, but in varying amounts; the amounts of TQ in the samples from Iraq, India, Saudi Arabia, Syria, and Turkey, respectively, were 0.031, 0.030, 0.022, 0.005, and 0.001%. According to a comparison of their performances, the four ensemble machine learning techniques can reproduce the chromatographic properties of TQ, PA, and tR with minimum and maximum NSE-values of 0.842 and 0.999 in the training phase and 0.918 and 0.999 in the testing phase, respectively. Conclusion: The TQ content of each sample of black cumin, which was collected from various geographical locations, was determined quantitatively. The quantity of thymoquinone fluctuates depending on geographic variances, according to HPLC data. Five different AI-based models, including SWLR, PCR, LSSVM, ANFIS, and LSQ-Boost, were used to simulate the chromatographic behavior of TQ information of retention duration and peak area using various independent factors. Additionally, SAE, WAE, NNE, and ANFIS-E are informed by the application of ensemble machine learning to enhance the performance of AI-based models. Comparing the approaches to the individual models, they both demonstrated lower error values in terms of RMSE and MSE.