Artificial intelligence-driven near-infrared spectrophotometry model for rapid quantification of anti-nutritional factors in soybean (Glycine max.)

Abstract

Anti-nutritional factors can impact soybean nutrient bioavailability when consumed by monogastric animals. However, conventional methods available for quantifying anti-nutritional factors such as phytate and trypsin inhibitors in feeds are laboratory-intensive, time-consuming, expensive, and error-prone. This study developed near-infrared spectrophotometry (NIRS)-based models to quantify phytate and trypsin inhibitors in soybean. Thus, a set of 190 soybean genotypes assayed through conventional wet chemistry was used as a reference for model development and cross-validation. Using a benchtop NIR instrument (DS2500), spectra readings between 400 and 2500 nm were taken from each soybean sample. Mean values for phytate and total trypsin inhibitors (TTI) were 1.77 mg g−1 (SD = 1.23) and 0.89 mg g−1 (SD = 0.24), respectively. Predictive models were developed through partial least squares (PLS) and random forest (RF) regressions. The random forest models outperformed partial least squares regression with the best predictive performance of R2test = 0.97; RPD = 5.95 and R2test = 0.96; RPD = 3.62 for phytate and TTI, respectively. The high R2 and RPD values demonstrate the model's strong predictive capability and accuracy, suggesting that the NIRS-based models can effectively quantify phytate and TTI in soybean. Thus, breeders can efficiently select for low-anti-nutritional genotypes and accelerate the development of nutritionally beneficial legumes while reducing soybean processing costs. NIRS offers a promising alternative to traditional phenotyping methods due to its speed, simplicity, environmental friendliness, and cost-effectiveness. Its integration into breeding programs can streamline the screening process, especially in early selection stages.