Quantitative structure-activity relationship modeling of alcohol, ester, aldehyde, and ketone flavor thresholds in beer from molecular features.

Three quantitative structure-activity relationship approaches-principal components regression, partial least-squares regression, and alternating conditional expectations-were used to investigate relationships between the flavor thresholds of 38 alcohols, 40 esters, 45 aldehydes, and 43 ketones in beer and their structures. Strong nonlinear relationships between the logarithm of the flavor threshold and four or five structure descriptors were found for each class of compounds (R2 = 0.920, 0.937, 0.920, and 0.928 for alcohols, esters, aldehydes, and ketones, respectively). Simple nonlinear relationships between the alcohol, ester, and aldehyde thresholds and the numbers of hydrogen atoms in the molecules were also demonstrated.