Machine learning screening of bile acid-binding peptides in a peptide database derived from food proteins

Bioactive peptides (BPs) are protein fragments that exhibit a wide variety of physicochemical properties, such as basic, acidic, hydrophobic, and hydrophilic properties; thus, they have the potential to interact with a variety of biomolecules, whereas neither carbohydrates nor fatty acids have such diverse properties. Therefore, BP is considered to be a new generation of biologically active regulators. Recently, some BPs that have shown positive benefits in humans have been screened from edible proteins. In the present study, a new BP screening method was developed using BIOPEP-UWM and machine learning. Training data were initially obtained using high-throughput techniques, and positive and negative datasets were generated. The predictive model was generated by calculating the explanatory variables of the peptides. To understand both site-specific and global characteristics, amino acid features (for site-specific characteristics) and peptide global features (for global characteristics) were generated. The constructed models were applied to the peptide database generated using BIOPEP-UWM, and bioactivity was predicted to explore candidate bile acid-binding peptides. Using this strategy, seven novel bile acid-binding peptides (VFWM, QRIFW, RVWVQ, LIRYTK, NGDEPL, PTFTRKL, and KISQRYQ) were identified. Our novel screening method can be easily applied to industrial applications using whole edible proteins. The proposed approach would be useful for identifying bile acid-binding peptides, as well as other BPs, as long as a large amount of training data can be obtained.