Efficient discovery of anticancer peptides via cost-aware ranking learning

While anticancer peptides (ACPs) have emerged as a promising class of next-generation therapeutics, efficiently identifying potent candidates across the vast peptide sequence space remains a pressing challenge. Here, we present ACPRank, a computational framework based on cost-aware ranking learning tailored for discovering potent ACPs. By integrating a unified activity scoring system encompassing multidimensional pharmacological metrics and employing a customized ranking loss, the framework directly embeds the core objective of targeting high-activity candidates under resource-limited conditions into the optimization process, thereby enabling precise candidate prioritization. Systematic screening of the human secretome facilitated the identification of two novel human-derived ACPs (termed DRP and KRP), both of which exhibited robust broad-spectrum antitumor efficacy in vitro and in vivo , alongside a favorable safety profile. Mechanistic investigations revealed that KRP induces marked cell cycle arrest by downregulating key mitotic regulators, including Aurora Kinase B (AURKB), Budding Uninhibited by Benzimidazoles 1 (BUB1), Cyclin B1 (CCNB1), and Non-SMC Condensin I Complex Subunit H (NCAPH). Overall, ACPRank enables rapid identification of highly active candidates and thus holds great potential for addressing the efficiency bottleneck in large-scale peptide screening.