Scholarlink

<p>COVID-19, as a novel disease, has limited treatment options, and the use of associated drugs may result in Adverse Drug Effects (ADE) that can lead to long-term health complications. Therefore, accurate prediction of ADE is critical in the development and evaluation of COVID-19 treatments. Existing studies primarily rely on traditional models for ADE prediction, underscoring the need for more effective computational approaches that can surpass their limitations. To address this, we propose the Chemical-ADE Co-Linker (CACL) classifier, a novel similarity-based method designed to predict ADE associated with COVID-19 drugs by leveraging chemical data, including 1D drug structures and 17 calculated molecular descriptors. The core concept of the model consists of two parts: (1) adverse drug events that commonly occur together indicate shared chemical characteristics among the linked pharmaceuticals, and (2) medications with similar chemical characteristics are likely to exhibit similar ADE. The CACL classifier operates in two stages: It first determines the top <i>N</i> chemically comparable drugs to a test substance by calculating similarity scores using training examples. The most common labels among these comparable medications are then used to give the ADE to the test drug. Second, the model refines these predictions by calculating ADE-ADE similarity using the Jaccard index based on co-occurrence patterns, selecting the top <i>M</i> most related ADE to finalize the prediction. Extensive evaluation shows that the CACL classifier achieves superior performance, reporting an accuracy of 98.26% on 1D chemical structure data. It outperforms classifiers from previous research studies, such as Random Forest, Multi-Layer Perceptron Neural Network, K1K2NN, Multi-Label Deep Neural Network, XGBoost, Extra Trees Classifier, K-Nearest Neighbors, AdaBoost, and Decision Tree.</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

Cacl: a novel chemical-ADE co-linker classifier for identifying adverse drug effects from chemical properties

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Pranab Das,
  • Sujit Kumar,
  • Nurul Amin Choudhury,
  • Bishal Chhetry,
  • Bam Bahadur Sinha,
  • Rajkumar Joydev Borah

摘要

COVID-19, as a novel disease, has limited treatment options, and the use of associated drugs may result in Adverse Drug Effects (ADE) that can lead to long-term health complications. Therefore, accurate prediction of ADE is critical in the development and evaluation of COVID-19 treatments. Existing studies primarily rely on traditional models for ADE prediction, underscoring the need for more effective computational approaches that can surpass their limitations. To address this, we propose the Chemical-ADE Co-Linker (CACL) classifier, a novel similarity-based method designed to predict ADE associated with COVID-19 drugs by leveraging chemical data, including 1D drug structures and 17 calculated molecular descriptors. The core concept of the model consists of two parts: (1) adverse drug events that commonly occur together indicate shared chemical characteristics among the linked pharmaceuticals, and (2) medications with similar chemical characteristics are likely to exhibit similar ADE. The CACL classifier operates in two stages: It first determines the top N chemically comparable drugs to a test substance by calculating similarity scores using training examples. The most common labels among these comparable medications are then used to give the ADE to the test drug. Second, the model refines these predictions by calculating ADE-ADE similarity using the Jaccard index based on co-occurrence patterns, selecting the top M most related ADE to finalize the prediction. Extensive evaluation shows that the CACL classifier achieves superior performance, reporting an accuracy of 98.26% on 1D chemical structure data. It outperforms classifiers from previous research studies, such as Random Forest, Multi-Layer Perceptron Neural Network, K1K2NN, Multi-Label Deep Neural Network, XGBoost, Extra Trees Classifier, K-Nearest Neighbors, AdaBoost, and Decision Tree.