Multivariate Calibration to Determine Phorbol Esters in Seeds of Jatropha curcas L. Using Near Infrared and Ultraviolet Spectroscopies

Abstract

The building of partial least squares (PLS) regression models using near infrared (NIR) and ultraviolet (UV) spectroscopies to estimate the concentrations of phorbol esters (PEs) in Jatropha curcas L. is presented. The models were built using two algorithms for variable selection, ordered predictors selection (OPS) and genetic algorithm (GA). Chromatographic analyses were performed to determine the concentrations of PEs. Spectral data were obtained from seeds and oil extract. The results of PLS models were performed by analyzing statistical parameters of quality such as root mean square error of prediction (RMSEP) and correlation coefficient of external predictions (Rp). The parameters obtained for NIR-PLS and UV-PLS models with OPS were respectively: RMSEP 0.48 and 0.22 mg g-1 and Rp 0.49 and 0.96. For GA were obtained, respectively: RMSEP 0.52 and 0.28 mg g-1 and Rp 0.12 and 0.95. The models built from seeds and oil extracts can be used respectively for screening and to accurately predict the PEs content. The OPS method provided simpler and more predictive models compared to those obtained by the selection of variables using the GA. Thus, the UV-PLS-OPS model can be used as an alternative method to quantification of PEs.