Scholarlink

<p>Pyrazole and pyrrolidine represent two classes of high-frequency nitrogen-containing “privileged scaffolds” in drug discovery and industrial chemicals; however, their potential acute toxicity poses significant challenges to clinical translation and industrial application. This study aims to establish systematic prediction models for the acute oral toxicity of the two scaffolds in rats and mice, strictly adhering to OECD guidelines. Based on experimental data for 552 compounds collected from PubChem, we calculated 2D molecular descriptors and systematically compared the performance of several modeling strategies, traditional 2D-QSTR, q-RASTR, Hybrid-ARKA, and ARKA-RASTR and six machine learning (ML) algorithms. The results demonstrated that the ARKA-RASTR framework yielded the most superior performance. It not only outperformed other methods in external validation but also effectively overcame the internal stability issues often associated with conventional q-RASTR approaches, with all validation metrics exceeding the most stringent international standards. In terms of mechanistic interpretation, this study innovatively employed the ARKA-RASTR model for intelligent physical mechanism analysis, dynamically linking variable importance to specific toxicity intensity ranges, thereby significantly enhancing model interpretability. Finally, the optimized models were applied to the virtual screening of 18,000 real world compounds lacking experimental values. Through applicability domain (AD) assessment, we provided prioritized lists of the top ten potential high- and low-toxicity candidates for each scaffold. We also developed an online web-based predictor: PPAOT (Pyrazole-Pyrrolidine-Acute Oral Toxicity), enabling one-stop toxicity prediction. By leveraging advanced data fusion strategies, this study offers robust tools and clear guidance for the early safety assessment and structural optimization of nitrogen-containing heterocyclic drugs and chemicals.</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

PPAOT predictor: ARKA-RASTR strategy-driven rodent acute oral toxicity prediction for pyrazole and pyrrolidine scaffolds-based chemicals with intelligent mechanistic interpretation

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Jianing Xu,
  • Shuo Chen,
  • Ting Ren,
  • Feifan Li,
  • Na Zhang,
  • Lijiao Zhao,
  • Rugang Zhong,
  • Guohui Sun

摘要

Pyrazole and pyrrolidine represent two classes of high-frequency nitrogen-containing “privileged scaffolds” in drug discovery and industrial chemicals; however, their potential acute toxicity poses significant challenges to clinical translation and industrial application. This study aims to establish systematic prediction models for the acute oral toxicity of the two scaffolds in rats and mice, strictly adhering to OECD guidelines. Based on experimental data for 552 compounds collected from PubChem, we calculated 2D molecular descriptors and systematically compared the performance of several modeling strategies, traditional 2D-QSTR, q-RASTR, Hybrid-ARKA, and ARKA-RASTR and six machine learning (ML) algorithms. The results demonstrated that the ARKA-RASTR framework yielded the most superior performance. It not only outperformed other methods in external validation but also effectively overcame the internal stability issues often associated with conventional q-RASTR approaches, with all validation metrics exceeding the most stringent international standards. In terms of mechanistic interpretation, this study innovatively employed the ARKA-RASTR model for intelligent physical mechanism analysis, dynamically linking variable importance to specific toxicity intensity ranges, thereby significantly enhancing model interpretability. Finally, the optimized models were applied to the virtual screening of 18,000 real world compounds lacking experimental values. Through applicability domain (AD) assessment, we provided prioritized lists of the top ten potential high- and low-toxicity candidates for each scaffold. We also developed an online web-based predictor: PPAOT (Pyrazole-Pyrrolidine-Acute Oral Toxicity), enabling one-stop toxicity prediction. By leveraging advanced data fusion strategies, this study offers robust tools and clear guidance for the early safety assessment and structural optimization of nitrogen-containing heterocyclic drugs and chemicals.