Discordance between in silico and in vitro results on proteolytic stability of milk protein-derived DPP-4 inhibitory peptides

The resistance of food-derived bioactive peptides to gastrointestinal enzymatic degradation remains a key challenge to address before considering their in vivo activity. While many in vitro studies explored peptide activity, proteolytic stability studies are limited. The proteolytic stability of some of the most potent proline-containing, milk protein-derived DPP-4 inhibitory peptides (IPI, IPA, IPM, LPQNIPPL, and INNQFLPYPY) from the BIOPEP-UWM database was initially screened using BIOPEP-UWM, ExPASy PeptideCutter, and ProsperousPlus in silico webservers. The peptides were synthesized and incubated separately in vitro with pepsin, chymotrypsin, or pancreatic elastase. The peptides were proteolytically stable during 3 h incubation, except INNQFLPYPY, which was partially degraded by chymotrypsin, supporting previous findings on hydrolysis resistance. However, the cleavage sites predicted by in silico tools were not fully confirmed by in vitro experiment. Misclassified cleavage sites were observed for proline-, asparagine-, or glutamine-containing peptide bonds, highlighting potential limitations of the commonly used in silico tools. To assess the observed differences, a statistical analysis was performed to evaluate predictive accuracy using a validation dataset of 897 previously tested cleavage sites. The in silico and in vitro agreement on cleavage sites was below 75%. Negative predictive values, supported by specificity data, indicate that in silico tools may miss 10–30% of the cleavage sites. The obtained likelihood ratio values (LR+ < 10, LR- > 0.1) did not meet the expected standards of accuracy. Analysis of the underlying methodology of the in silico tools clarified the potential causes of conflicting results and identified areas for improvement.