Uncertainty profile for NIR analysis of tanshinone I content in tanshinone extract powders

A rapid, non-destructive and reliable analytical method using NIR diffuse reflectance spectroscopy combined with variable selection methods was developed and validated for the quantification of tanshinone I content in tanshinone extract. After spectral pretreatment, different variables selection methods such as interval partial least square (iPLS), synergy interval partial least square (SiPLS), uninformative variables elimination (UVE), successive projections algorithm (SPA) and competitive adaptive reweighted sampling (CARS) were used to select sensitive variables and to develop partial least squares (PLS) regression models. Results indicated that UVE method was an efficient method to eliminate the redundant information, extract useful features and improve model performance. The root mean squared error of prediction (RMSEP) and ratio of performance to deviation (RPD) of the quantitative model were improved form 0.483% and 12.26 to 0.433% and 13.67, respectively. A global strategy was proposed to examine the validity of the built NIR method as well as to estimate its uncertainty at the same time. And an innovative formula we offered to assess the uncertainty was based on the calculation of the β-content tolerance interval by the Hoffman-Kringle approach. Furthermore, a novel approach based on uncertainty profile (UP) was used to validate the robustness and accuracy of PLS model. It concluded that NIR analysis combined with variables selection method was suitable and reliable for quantification of tanshinone I content in tanshinone extract, and could be applied to control the quality of tanshinone extract and provide reference for quality control of other Chinese herbal medicines.