Abstract
Background: Niacin (NIA) is a water-soluble vitamin and the primary treatment of pellagra. No analytical method was found to assess NIA in complex mixtures with its official impurities.
Objective: Two validated, accurate, and selective chemometric models were developed to assay NIA in the presence of its four official impurities, including pyridine, a nephrotoxic and hepatotoxic substance. Additionally, the two selective chemometric models were compared by processing UV spectra in the range 220-305 nm and applying partial least squares regression (PLSR) and support vector regression (SVR) models.
Method: A five levels five factors experimental design was chosen to exhibit a training set of 25 mixtures that had numerous variable percentages of tested substances. A test set consisting of 10 mixtures was designed to confirm the predictive power of the suggested models.
Results: The presented results substantiate the strength of the developed multivariate calibration models to assay NIA specifically with high selectivity and accuracy (100.02 +/- 1.312 and 100.04 +/- 1.272 for PLSR and SVR models, respectively). The root mean square error of prediction for the validation set mixtures was applied as a main comparison tool and it was found to be 0.2016 and 0.1890 for PLSR and SVR models, respectively.
Conclusions: The results of the developed models and the reported HPLC method were statistically compared, where F-values and Student's t-tests did not show significant difference in regards to accuracy and precision. The SVR model proved to be more accurate than the PLSR model, producing a high generalization capacity, while PLSR was easy to implement and fast.
Highlights:
Assay of Niacin with its official impurities.
Comparative study of PLSR and SVR models.
Coupling of UV data and chemometrics for pharmaceutical analysis.