Human milk is widely regarded as the gold standard of infant nutrition, representing a complex mixture of vital nutrients and bioactive components. Among these, milk-derived peptides, generated during protein digestion, play important roles in many physiological processes such as antimicrobial activity, antihypertensive regulation, and metabolic modulation. Linking such different biological functions to distinct, identifiable regions within peptides’ chemical space remains however an open challenge. In this study, we analysed approximately 91k annotated bioactive peptides to investigate the strength of structure-property relationships and to assess the possibility of discriminating peptide function based on their chemical characteristic, using computational approaches. To this end, we started from feature engineering, where two complementary molecular representations were developed: the first consisted of physicochemical descriptors, capturing interpretable properties, such as charge-related features, topological indices, and fingerprint-derived measures.