Scholarlink

<p>Antimicrobial resistance (AMR) is a pressing global health challenge, particularly for pathogens like <i>Escherichia coli</i> (<i>E. coli</i>), a WHO Bacterial Priority Pathogen known for resistance to multiple antibiotics, including third-generation cephalosporins. Developing novel antibiotics is resource-intensive, making combination therapy an attractive strategy to enhance efficacy and mitigate resistance through synergistic drug interactions. Identifying effective combinations is complex due to the vast number of possible drug pairs and their interactions across diverse biological contexts. We developed a machine learning (ML) framework integrating drug structure information, pathogen response data, and disease microenvironment data to predict synergistic drug combinations for <i>E. coli</i>. Using a dataset obtained from the literature comprised of 392 drug combinations (314 and 78 for training and validation, respectively), having 4145 features (3979 <i>E. coli</i> genes and 166 substructural profiles), we trained a suite of machine learning models including decision tree (DT), random forest (RF), support vector machine (SVM), and gradient boost classifier (GBC). For the test set, we leveraged an external dataset consisting 44 drug combinations with known interaction scores but not used during training or validation. Among single classifiers, GBC achieved the best performance, with 72% ROC-AUC, 70% accuracy, 66% precision, 68% recall, and 67% F1 score. From a dataset of 306 antibiotic combinations with unknown interaction scores, GBC predicted 37 synergistic drug combinations across two media, with metabolic conditions significantly influencing outcomes (e.g., amikacin-tetracycline was synergistic in glucose but antagonistic in glycerol). Feature importance analysis highlighted key genes like <i>oxyR</i> and pathways such as amino acid biosynthesis. This integrated approach demonstrates the potential of ML to accelerate the discovery of effective combination therapies, providing a scalable framework that could be applied to other resistant pathogens, emphasizing the need for cross-species validation and experimental testing of candidate combinations to support clinical translation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Discovery of synergistic drug combinations for E. coli from drug information, pathogen response and disease microenvironment data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月3日

查看全文
  • Racheal Claire Kyomukama,
  • Moses Ainembabazi,
  • Henry Mutegeki,
  • Patricia Nabisubi,
  • Fredrick Elishama Kakembo,
  • Gloria Loyce Nankoma,
  • Kevin Cissy Nabukeera,
  • Andrew Walakira,
  • Ronald Galiwango

摘要

Antimicrobial resistance (AMR) is a pressing global health challenge, particularly for pathogens like Escherichia coli (E. coli), a WHO Bacterial Priority Pathogen known for resistance to multiple antibiotics, including third-generation cephalosporins. Developing novel antibiotics is resource-intensive, making combination therapy an attractive strategy to enhance efficacy and mitigate resistance through synergistic drug interactions. Identifying effective combinations is complex due to the vast number of possible drug pairs and their interactions across diverse biological contexts. We developed a machine learning (ML) framework integrating drug structure information, pathogen response data, and disease microenvironment data to predict synergistic drug combinations for E. coli. Using a dataset obtained from the literature comprised of 392 drug combinations (314 and 78 for training and validation, respectively), having 4145 features (3979 E. coli genes and 166 substructural profiles), we trained a suite of machine learning models including decision tree (DT), random forest (RF), support vector machine (SVM), and gradient boost classifier (GBC). For the test set, we leveraged an external dataset consisting 44 drug combinations with known interaction scores but not used during training or validation. Among single classifiers, GBC achieved the best performance, with 72% ROC-AUC, 70% accuracy, 66% precision, 68% recall, and 67% F1 score. From a dataset of 306 antibiotic combinations with unknown interaction scores, GBC predicted 37 synergistic drug combinations across two media, with metabolic conditions significantly influencing outcomes (e.g., amikacin-tetracycline was synergistic in glucose but antagonistic in glycerol). Feature importance analysis highlighted key genes like oxyR and pathways such as amino acid biosynthesis. This integrated approach demonstrates the potential of ML to accelerate the discovery of effective combination therapies, providing a scalable framework that could be applied to other resistant pathogens, emphasizing the need for cross-species validation and experimental testing of candidate combinations to support clinical translation.