Scholarlink

<p>Antimicrobial resistance (AMR) is one of the main global health concern, and it leads to the consideration of rapid and scalable computational methodologies of microorganism analysis. This paper proposes an attention-based deep learning framework of morphology-based microorganism classification and exploratory phenotype-associated pattern analysis using a publicly available dataset. Various deep learning models such as CNNs, RNNs, LSTMs, GRUs, FCNNs, MLPs, attention-based and hybrid models are tested to examine their discriminative morphological-representation learning capability. The dataset is pre-processed to combat the level of imbalance in the classes and the variability of the features. The attention-based CNN has been found to exhibit the highest level of generalization with training and validation accuracies of 99.78% and 99.49 respectively. The architecture of hybrid architectures is also competitive, and the FCNN-RNN model achieves a validation accuracy of 98.87%. Attention-based and hybrid models invariably outperform single-architecture systems in AUC-ROC, indicating the usefulness of attention mechanisms in classifying morphology-based microorganisms. Notably, a direct prediction of antibiotic resistance is not done in this study because there are no molecular resistance markers or antimicrobial susceptibility labels. The proposed framework conducts morphology-based classification of microorganisms, and exploratory analysis of patterns that can be associated with a phenotype in the curated dataset and the results reported do not reflect a prediction of antimicrobial resistance.</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

Attention-based deep learning for morphology-based microorganism classification and exploratory phenotype pattern analysis

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • S J K Jagadeesh kumar,
  • G. Prabu kanna,
  • Inderpreet Kaur,
  • Yogesh Kumar

摘要

Antimicrobial resistance (AMR) is one of the main global health concern, and it leads to the consideration of rapid and scalable computational methodologies of microorganism analysis. This paper proposes an attention-based deep learning framework of morphology-based microorganism classification and exploratory phenotype-associated pattern analysis using a publicly available dataset. Various deep learning models such as CNNs, RNNs, LSTMs, GRUs, FCNNs, MLPs, attention-based and hybrid models are tested to examine their discriminative morphological-representation learning capability. The dataset is pre-processed to combat the level of imbalance in the classes and the variability of the features. The attention-based CNN has been found to exhibit the highest level of generalization with training and validation accuracies of 99.78% and 99.49 respectively. The architecture of hybrid architectures is also competitive, and the FCNN-RNN model achieves a validation accuracy of 98.87%. Attention-based and hybrid models invariably outperform single-architecture systems in AUC-ROC, indicating the usefulness of attention mechanisms in classifying morphology-based microorganisms. Notably, a direct prediction of antibiotic resistance is not done in this study because there are no molecular resistance markers or antimicrobial susceptibility labels. The proposed framework conducts morphology-based classification of microorganisms, and exploratory analysis of patterns that can be associated with a phenotype in the curated dataset and the results reported do not reflect a prediction of antimicrobial resistance.