Prediction of glucose, fructose and sucrose content in cassava (Manihot esculenta Crantz) genotypes from Amazon using PLS models

Abstract

The chemical characterization by classical methods requires a long time of analysis and the use of expensive and environmentally aggressive reagents. The use of the partial least squares (PLS) tool applied to FT-MIR data represents a reduction of these considered variables. The relative contributions of glucose, fructose, and sucrose obtained for the 26 cassava samples varied between 0.111-0.383 g/100g; 0.0317-0.256 g/100g and 0.286-0.775 g/100g, respectively. For five latent variables the mean of predicted glucose content in external samples was 0.220 g/100g and had the RMSEP value of 0.00590 g/100g; The best number of LVs for the prediction of the fructose content for new samples were five, where the mean of the predicted value was 0.0994 g/100g against the mean fructose reference value 0.0879 g/100g, with a 0.0115 g/100g RMSEP; The mean sucrose content in the external samples was 0.451 g/100g, compared with the reference value 0.515 g/100g, with a RMSEP 0.138 g/100g. The use of the PLS1 algorithm generated two good models with the second derivative in spectral data and one with the raw data in spectral data using four samples external to the prediction step. The mid infrared spectrum shows a facile method to quantify glucose, fructose and sucrose obtaining PLS models with considerable robustness.