Non-destructive genotypes classification and oil content prediction using near-infrared spectroscopy and chemometric tools in soybean breeding program

In soybean (Glycine max L.) breeding programs, segregation is normally observed, and it is not possible to have replicates of individuals because each genotype is a unique copy. Therefore, near-infrared spectroscopy (NIRS) was used as a non-destructive tool to classify soybeans by genotypes and to predict oil content. A total of 260 soybean genotypes were divided into five classes, which were composed of 32, 52, 82, 46, and 49 samples of the BV, BVV, EB, JAB, and L class, respectively. NIR spectra were obtained using oven-dried samples (80 g) in a reflectance mode. A successive projection algorithm and genetic algorithm with linear discriminant analysis discriminated genotypes of the low (L class) from the high (EB class) for oil content (88.89% accuracy). The partial least square regression models for oil content were considered good (root mean square error of prediction of 0.96%). Therefore, NIRS can be used as a non-destructive tool in soybean breeding programs, but further investigation is necessary to improve the robustness of the models. It is important to note that to use the models, it is necessary to collect NIR spectra from dry soybean samples.