Rapid ATR-FTIR-Based Prediction of Barley Malt Quality

Barley malt quality analysis is essential for craft malt houses, commercial producers, and barley breeding programs, yet traditional methods are resource-intensive and time-consuming. We combined attenuated total reflectance-Fourier transform infrared (ATR-FTIR) spectroscopy with automated multivariate regression models to predict eight malt quality traits. Using super partial least-squares (sPLS) and super support vector regression (sSVR), quality metrics were evaluated using 460 barley lines from three U.S. locations. External validation coefficients of determination (Rv2) were the highest for barley protein (BP), soluble protein/total protein (S/T), and malt extract (ME). These values were respectively 0.946, 0.907, and 0.880 for sSVR and 0.942, 0.901, and 0.867 for sPLS. Average validation errors for ME, BP, and S/T were 1.3%, 4.6%, and 6.3%, respectively, for sSVR and 1.4%, 4.7%, and 6.5% for sPLS, respectively. Variable selection methods─binning, wavelet, and variance feature ranking (VFR)─were compared, with binning and wavelet techniques retaining information effectively even when measurements were reduced by 95%. The ratio of performance to deviation (RPD) values of BP, S/T, and ME models for sPLS to sSVR were 4.2 to 4.3, 3.2 to 3.3, and 2.7 to 2.9, respectively. These models provide an efficient, nondestructive alternative for rapid barley malt quality analysis, offering scalable tools for research and industry applications in barley breeding and malt production.