Nature子刊：计剑/黄俊杰等开发小样本深度学习模型，高效、精准挖掘鲍曼不动杆菌抗菌肽

Core Viewpoint - The research highlights the urgent need for new antimicrobial agents targeting Acinetobacter baumannii due to its high mortality rate and increasing antibiotic resistance, with a focus on the development of antimicrobial peptides (AMPs) as a promising alternative to traditional antibiotics [2][3][17]. Group 1: Antimicrobial Resistance and Research Background - Acinetobacter baumannii is recognized by the WHO as a critical priority for the development of new antibiotics due to its high clinical infection mortality rate and growing resistance to existing treatments [2]. - Polymyxin is currently used to treat infections caused by Acinetobacter baumannii but is associated with nephrotoxicity and has seen rising bacterial resistance [2]. Group 2: Development of Antimicrobial Peptides - Antimicrobial peptides are considered ideal candidates for next-generation antibiotics due to their mechanism of action, which disrupts bacterial cell membranes and is less likely to induce resistance [2]. - The research team developed a novel workflow called FSLSMEP, which integrates pre-training and fine-tuning techniques to address the scarcity of data on antimicrobial peptides targeting Acinetobacter baumannii [3][9]. Group 3: Methodology and Results - The FSLSMEP workflow utilizes a pre-trained model (ESM) and fine-tunes it with similar data from Pseudomonas aeruginosa antimicrobial peptides, significantly enhancing predictive capabilities [8][9]. - The workflow successfully identified potent antimicrobial peptides from a theoretical library of over 256 billion peptides, achieving a 91.1% success rate in demonstrating antimicrobial activity [9][17]. Group 4: Mechanism of Action and Efficacy - The study confirmed that the antimicrobial peptides disrupt bacterial cell membranes, supported by electron microscopy and RNA-seq results, indicating a low likelihood of developing resistance [10][17]. - EME7(7), one of the identified peptides, exhibited excellent antimicrobial performance with a minimum inhibitory concentration (MIC) of 8 μg/mL and demonstrated superior in vivo efficacy compared to polymyxin B, with no nephrotoxicity observed [9][10][17]. Group 5: Generalizability and Future Applications - The FSLSMEP workflow was also successfully applied to discover antimicrobial peptides against Candida albicans, demonstrating its versatility in addressing data scarcity across different pathogens [13][17]. - This research provides a new paradigm for tackling the challenge of limited data in antimicrobial peptide discovery, potentially leading to effective treatments for resistant bacterial infections [17].