Discovery of antimicrobial peptides targeting Acinetobacter baumannii via a pre-trained and fine-tuned few-shot learning-based pipeline

Acinetobacter baumannii, a robust Gram-negative bacterium known for causing nosocomial infections, exhibiting multidrug resistance, and lacking antimicrobial peptides that target it, remains hard to treat. Here, we report a few-shot learning pipeline integrating classification, ranking, and regression modules. Each module is trained via a few-shot learning strategy involving pre-training and multiple fine-tuning steps, incorporating similar and true data for fine-tuning, to identify potent AMPs against Acinetobacter baumannii. This pipeline effectively scans complete libraries of hexapeptides, heptapeptides, and octapeptides (encompassing tens of billions of candidates) despite the extreme scarcity of training data. Results show it discovers AMPs active against Acinetobacter baumannii and Candida albicans, with low off-target toxicity and negligible drug resistance susceptibility. Additionally, EME7(7) controls Acinetobacter baumannii pneumonia in mice without kidney injury, a contrast to the observed effects of polymyxin B. This work provides a paradigm for addressing challenges of limited data availability. The authors introduce a deep learning pipeline integrating classification, ranking, and regression modules, in which each module is trained via a few-shot learning strategy involving pretraining and multiple fine-tuning steps, to identify potent AMPs against Acinetobacter baumannii.