The multivariate data analysis software The Unscrambler v.9.8 (Camo, Oslo, Norway) was used to perform the statistical analysis. Mangosteen samples were divided into calibration (n = 81, which represented 60% of the whole sample set) and prediction (n = 54) sets. Both sets covered identical variations and uniform distribution of the predictor values (Table 1). Partial least squares regression (PLSR) models were built for each predictor variable with the calibration set. The optimal number of PLSR factors which leads to the lowest standard error of cross validation (SECV) was determined by full cross validation with the calibration sample. The performance of the models was validated on the prediction set by comparison of the measured values and model estimated values. The statistics for assessment of performance consisted of the correlation coefficient of prediction (rp) and the root mean square error of prediction (RMSEP)