Classification-based QSAR Models for the Prediction of the Bioactivity of ACE-inhibitor Peptides

Background: Local classification models were used to establish Quantitative Structure-Activity Relationships (QSARs) of bioactive di(-), tri(-) and tetrapeptides, with their capacity to inhibit Angiotensin Converting Enzyme (ACE). These discrete models can thus predict this activity for other peptides obtained from functional foods. These types of peptides allow some foods to be considered nutraceuticals. Method: A database of 313 molecules of di(-), tri(-) and tetrapeptides was investigated and antihypertensive activities of peptides, expressed as log (1/IC50), were separated into two qualitative classes: low activity (inactive) was associated with experimental values under the 66th percentile and active peptides with values above this threshold. Chemicals were divided into a training set, including 70% of the peptides, and a test set for external validation. Genetic algorithms-variable subset selection coupled with the kNN and N3 local classifiers were applied to select the best subset of molecular descriptors from a pool of 953 Dragon descriptors. Both models were validated on the test peptides. Results: The N3 model turned out to be superior to the kNN model when the classification focused on identifying the most active peptides.