Deep Learning Unravels Potential Antibiotic-Resistance Drugs against Klebsiella Pneumoniae

Abstract

BackgroundKlebsiella pneumoniae (KP) is a critical pathogen recognized by the World Health Organization due to its high mortality rates and resistance to antibiotics. Developing effective treatments for KP is essential to mitigate the growing threat of multidrug-resistant infections. This study leverages deep learning techniques to identify and validate potential drug candidates against KP, accelerating the drug discovery process and reducing associated risks. ResultsA comprehensive dataset was constructed, comprising 3,475 drugs selected from the DrugBank database, filtered for their potential efficacy and safety. Using convolutional neural networks (CNNs) for drug-drug interaction analysis, a prediction accuracy of 72% was achieved. Evolutionary Scale Modeling (ESM) was employed to calculate molecular similarities between drugs and KP strains, identifying five promising candidates with structural similarities exceeding 85% to known KP targets. A drug repurposing knowledge graph (DRKG) further validated these findings, with the PairRE model outperforming other approaches in ranking potential drug candidates. ConclusionsThis study presents a scalable and efficient framework for identifying and validating drug candidates to combat antibiotic-resistant KP. By integrating deep learning models, molecular similarity analysis, and knowledge graph techniques, this approach shortens the drug discovery timeline from years to months. The identified candidates hold significant promise for preclinical and clinical testing, offering a pathway to effective therapies against KP and other multidrug-resistant pathogens.