<p>Peptide quantitative structure-activity relationship (PepQSAR) has attracted much attention in the bio- and cheminformatics communities as a well-established computational peptidology strategy to statistically correlate the sequence/structure and activity/function of bioactive peptides (BAPs). In this study, a new concept termed DeepPepQSAR that integrates deep learning into traditional PepQSAR is proposed to quantitatively model, predict, and interpret the BAP universe in an all-in-one manner, that is, massive BAP samples with diverse activity types (i.e. antimicrobial, antiviral, hemolytic, anticancer, antigen, ACE-inhibitory, antioxidant, domain-binding, etc.) are merged into a single all-in-one DeepPepQSAR framework for artificial intelligence (AI)-driven big-data BAP discovery. A novel PepImage map is described to graphically represent both the sequence features of length-varying peptides and the activity types tested for these peptides, which is then fed into a dual-path, single-/multiple-channel convolutional neural network (CNN) for training, developing, and validating DeepPepQSAR regression models. We also practice the CNN-based DeepPepQSAR methodology on extrapolative navigation across a large-scale molecular diversity space covering billions of peptidic fragment candidates generated systematically from various food-derived proteins (FDPs) for AI-driven antimicrobial food peptide (AMFP) and antihypertensive food peptide (AHFP) discovery. Consequently, 14 AMFP peptides and 10 AHFP peptides are determined to have good antibacterial and ACE-inhibitory profiles, in which 4 and 2 peptides exhibit high potencies, respectively.</p> Graphical abstract <p></p>