Scholarlink

<p>Antimicrobial resistance poses a major global health challenge, particularly in infections involving <i>Staphylococcus aureus</i> and <i>Candida albicans</i>, which frequently occur together and exhibit synergistic pathogenicity. Rapid and efficient discovery of novel antimicrobials is essential and must align with sustainable, green chemistry practices. Here, we present a hybrid machine learning (ML) workflow for the simultaneous prediction of the logarithm of the minimal inhibitory concentration (ln(MIC)) of 126 bis-imidazolium chlorides against both pathogens. From 5,270 molecular descriptors, a multistage reduction pipeline—including redundancy filtering, classification and regression tree analysis, and principal component analysis—yielded twelve principal components explaining 91.7% of the variance. These were used as inputs for an ensemble of multilayer perceptron artificial neural networks (12-11-2 architecture, tanh activation for both hidden and output layers), enhancing predictive stability and generalizability. The resulting models achieved correlation coefficients of <i>r</i> = 0.966 for <i>S. aureus</i> and <i>r</i> = 0.954 for <i>C. albicans</i>, with ensemble modeling particularly improving the fungal predictions. This study demonstrates that even with limited data, a carefully designed hybrid ML approach can provide accurate, dual-target antimicrobial predictions, offering a computationally efficient and interpretable tool to accelerate antimicrobial screening while supporting sustainable drug discovery.</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

Hybrid Machine Learning Model for Dual Prediction of Antimicrobial Activity against Staphylococcus aureus and Candida albicans

  • 期刊论文

  • 发表日期：

    2026年2月5日

查看全文
  • Andżelika Lorenc,
  • Anna Badura,
  • Maciej Karolak,
  • Łukasz Pałkowski,
  • Łukasz Kubik,
  • Andrzej Skrzypczak,
  • Adam Buciński

摘要

Antimicrobial resistance poses a major global health challenge, particularly in infections involving Staphylococcus aureus and Candida albicans, which frequently occur together and exhibit synergistic pathogenicity. Rapid and efficient discovery of novel antimicrobials is essential and must align with sustainable, green chemistry practices. Here, we present a hybrid machine learning (ML) workflow for the simultaneous prediction of the logarithm of the minimal inhibitory concentration (ln(MIC)) of 126 bis-imidazolium chlorides against both pathogens. From 5,270 molecular descriptors, a multistage reduction pipeline—including redundancy filtering, classification and regression tree analysis, and principal component analysis—yielded twelve principal components explaining 91.7% of the variance. These were used as inputs for an ensemble of multilayer perceptron artificial neural networks (12-11-2 architecture, tanh activation for both hidden and output layers), enhancing predictive stability and generalizability. The resulting models achieved correlation coefficients of r = 0.966 for S. aureus and r = 0.954 for C. albicans, with ensemble modeling particularly improving the fungal predictions. This study demonstrates that even with limited data, a carefully designed hybrid ML approach can provide accurate, dual-target antimicrobial predictions, offering a computationally efficient and interpretable tool to accelerate antimicrobial screening while supporting sustainable drug discovery.