Antimicrobial Peptides Against Antimicrobial-Resistant Bacteria: Focus on Machine Learning

Le résumé fourni par la source

Abstract: Antimicrobial resistance (AMR) represents a pressing public health threat of the 21st century, with an estimated ten million deaths annually from drug-resistant infections by 2050. Diminishing pipelines and the accelerating emergence of multidrug-resistant pathogens make the development of novel antibacterials more urgent than ever. Antimicrobial peptides (AMPs) are among the most promising alternatives to conventional drugs, exhibiting broad antimicrobial spectra, rapid kinetics, and mechanisms that are difficult for bacteria to circumvent. However, the problem of discovering and engineering clinically useful AMPs with desirable properties out of large sequence spaces remains unsolved by traditional approaches. Machine learning (ML) enables fast screening of millions of compounds, generation of de novo sequences with predicted therapeutic potential, and simultaneous multiobjective optimisation of efficacy, safety, stability, and manufacturability. This review provides a critical appraisal of the current advances and prospective directions in computational discovery of AMPs that can combat resistant strains, focusing on available resources for machine learning in the domain of bioinformatics, evaluation of existing approaches to modeling peptide structure, activity, and interactions ranging from classical ML algorithms to DL and generative artificial intelligence (AI) models, and a practical roadmap of how the AMP discovery pipeline could proceed towards animal studies and clinical application through the use of active learning, fine-tuned protein language models, structural graph neural networks, and other modern techniques. Finally, we discuss challenges that may hinder a successful transition from ML-assisted design to the clinic and offer actionable recommendations to overcome them. Comparison of traditional and AI methods for antimicrobial peptide synthesis.The image compares two methods for antimicrobial peptide synthesis. The traditional method involves scientists using lab equipment to select molecules, resulting in high attrition. The text reads: ′Molecules picked by traditional methods for antimicrobial peptide synthesis.′ In contrast, the predictive AI machine learning model uses AI technology to select molecules, leading to low attrition. The text reads: ′Molecules picked by predictive AI model for antimicrobial peptide synthesis.′ The comparison highlights the efficiency of AI over traditional methods. Keywords: antimicrobial peptides, antimicrobial resistance, machine learning, DL, protein language models, drug discovery, clinical trials, geometric DL

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.