Validation of near-infrared spectroscopy for the quantification of cafestol and kahweol in green coffee

Abstract

Near-infrared spectroscopy (NIRS) is among the many tools available to study the biochemical diversity of coffee species. This technique is inexpensive, fast and accurate, and it requires only small amounts of samples. The aim of this study was to use NIRS to estimate the amount of diterpenes (cafestol and kahweol) in green coffee. To construct the prediction model, 126 Ethiopian accessions coffee collection and 44 modern cultivars were analyzed. The total sample set was split into two groups as follows: a group of 130 samples for calibration and a group of 40 samples for the validation step. Reference values of cafestol and kahweol were determined by high performance liquid chromatography (HPLC). Cafestol values ranged from 182.62g to 1308.62mg 100g−1, and kahweol values ranged from 182.69 to 1265.41mg 100g−1. To improve the quality of the calibration step, a pretreatment with the second derivative was applied to smooth the raw spectra. The prediction models of cafestol and kahweol were developed using the modified partial least squares regression (mPLS). The performance of these models was evaluated by the ratio of performance deviation (RPD) and R2 parameters, obtained by the ratio of the NIR prediction data and the corresponding reference data. The prediction models of cafestol (RPD=2.74; R2=0.89) and kahweol (RPD=2.2; R2=0.88) confirm the validity of NIRS analysis to determine diterpenes contents in green coffee.