Development of in silico prediction model for standardized ileal digestibility of amino acids from protein sources

Digestibility is an important factor in the nutritional evaluation of proteins because it determines the bioavailability of the amino acids derived from them. In this study, we developed a prediction model for standardized ileal digestibility (SID) of amino acids using previously reported pig SID datasets. The prediction was performed using the random forest model with the explanatory variables, including nutrients, protein types, cooking/processing, and other ingredients. Shapley additive explanations and feature importance analysis identified the factors of SID prediction and the correlation between digestibility and nutrient data. The predictability for pig SID data was compared between the prediction model and the reported digestibility data of an existing alternative method– in vitro INFOGEST digestion protocol. Finally, the digestible indispensable amino acid score (DIAAS) evaluation using predicted SID showed high accuracy compared with that using observed SID.