Venomics AI: a computational exploration of global venoms for antibiotic discovery.

The relentless emergence of antibiotic-resistant pathogens, particularly Gram-negative bacteria, highlights the urgent need for novel therapeutic interventions. Drug-resistant infections account for approximately 5 million deaths annually, yet the antibiotic development pipeline has largely stagnated. Venoms, representing a remarkably diverse reservoir of bioactive molecules, remain an underexploited source of potential antimicrobials. Venom-derived peptides, in particular, hold promise for antibiotic discovery due to their evolutionary diversity and unique pharmacological profiles. In this study, we mined comprehensive global venomics datasets to identify new antimicrobial candidates. Using machine learning, we explored 16,123 venom proteins, generating 40,626,260 venom-encrypted peptides (VEPs). Using APEX, a deep learning model combining a peptide-sequence encoder with neural networks for antimicrobial activity prediction, we identified 386 VEPs structurally and functionally distinct from known antimicrobial peptides. Our analyses showed that these VEPs possess high net charge and elevated hydrophobicity, characteristics conducive to bacterial membrane disruption. Structural studies revealed considerable conformational flexibility, with many VEPs transitioning to α-helical conformations in membrane-mimicking environments, indicative of their antimicrobial potential. Of the 58 VEPs selected for experimental validation, 53 displayed potent antimicrobial activity. Mechanistic assays indicated that VEPs primarily exert their effects through bacterial membrane depolarization, mirroring AMP-like mechanisms.