Predictive Model of Soymilk Sweetness Based on Protein and Amino Acid Compositions of Soybean Materials

In this study,sweetness values of soymilk made from 30 varieties of soybeans were determined by electronic tongues. Correlation analysis was used to explore the relationship between soymilk sweetness value and protein as well as amino acid compositions of soybean raw materials. Stepwise regression was applied to build soymilk sweetness predictive model. Results indicated that great variations of protein and amino acid compositions exist in different varieties of soybeans. Glycinin(11S)contents(r=0.370),glycinin/β-conglycinin ratio(11S/7S ratio)(r=0.436),serine contents(r=0.418)and threonine contents(r=0.373)were significantly positively correlated with soymilk sweetness(p<0.05). The contents of α subunit(r=-0.460),β-conglycinin(7S)(r=-0.428),methionine(r=-0.372)and tyrosine(r=-0.464)were significantly negatively correlated with soymilk sweetness(p<0.05). Predictive model of soymilk sweetness established by stepwise regression had a coefficient of determination R²=0.747,the equation was:F(predicted sweetness)=-0.125×α subunit+3.172×threonine+1.655×serine-2.894×methionine-2.097×tyrosine+9.908. An average relative error of 4.61% was obtained for predicted values compared with measured values. Therefore,the predictive model would have demonstrated great potential in accurately predicting soy milk sweetness values.