Scholarlink

This study introduces a virtual screening framework that combines multimodal molecular features with a stacking ensemble approach to identify potential HDAC6 inhibitors. A dataset of 1,701 compounds from ChEMBL33 and BindingDB was preprocessed and normalized, and adaptive sample reweighting via MetaWeightNet was applied to mitigate overfitting. The ensemble model, which includes XGBoost classifier arranged in two levels with a meta-learner, obtained an accuracy of 90.69%, a F1-score of 87.38%, an AUC of 94.93%, and a recall near 90%. Using this model to screen roughly 91 million compounds from PubChem led to 18 candidates, among which 17 exhibited docking scores lower than the reference compound SAHA (−7.8 kcal/mol), suggesting potential inhibitory activity. This framework offers a reliable approach for prioritizing compounds for experimental validation, supporting early-stage drug discovery by improving efficiency while reducing computational and experimental costs.

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

A Novel Multimodal Feature and DL-Based Virtual Screening Approach for HDAC6 Inhibitors Discovery

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nguyen Thanh Son,
  • Do Thi Mai Dung,
  • Le Van Tue,
  • Vu Hoang Thuy Duong,
  • Nguyen Ngoc An

摘要

This study introduces a virtual screening framework that combines multimodal molecular features with a stacking ensemble approach to identify potential HDAC6 inhibitors. A dataset of 1,701 compounds from ChEMBL33 and BindingDB was preprocessed and normalized, and adaptive sample reweighting via MetaWeightNet was applied to mitigate overfitting. The ensemble model, which includes XGBoost classifier arranged in two levels with a meta-learner, obtained an accuracy of 90.69%, a F1-score of 87.38%, an AUC of 94.93%, and a recall near 90%. Using this model to screen roughly 91 million compounds from PubChem led to 18 candidates, among which 17 exhibited docking scores lower than the reference compound SAHA (−7.8 kcal/mol), suggesting potential inhibitory activity. This framework offers a reliable approach for prioritizing compounds for experimental validation, supporting early-stage drug discovery by improving efficiency while reducing computational and experimental costs.