Determination of Total Sugar Content in Soy-Based Drinks Using Infrared Spectroscopy and Chemometrics

Partial least square (PLS) regression models were developed and compared in order to determine the total sugar content in soy-based drinks using an infrared spectroscopy technique known as attenuated total reflectance Fourier transform infrared (ATR-FTIR). On a spectrophotometer set for analyzing on the middle infrared region, spectral band of 1900 to 900 cm−1, commercial samples of soy beverage were analyzed, as well as samples with crescent water additions of 5, 10, and 20% v/v. Reference data for total sugars were obtained using the Lane-Eynon method. To construct regression models, algorithms of interval partial least square (iPLS) and synergy of interval partial least square (siPLS) were applied using iToolbox package on Matlab 8.1 environment. Kennard-Stone algorithm was used to the selection of calibration and prediction sets. Two models have been the best obtained: the first was an iPLS with seven latent variables, which selected the spectral band of 1399–900 cm−1 and presented root mean square error of cross-validation (RMSECV) = 0.1678% (w/w). The second best model was siPLS with six latent variables, which selected spectral bands of 1025–1150 and 1151–1476 cm−1 and presented RMSECV = 0.1963% (w/w). The proposed method presents advantages such as a small-required amount of sample for spectrum achievement, no sample destruction, and a high analytical frequency.