Scholarlink

<p>DNA polymerase theta (Polθ or POLQ) is an attractive target for treating BRCA-deficient cancers. In the present work, several computational approaches were employed for the design and discovery of novel POLQ helicase domain inhibitors. A dataset was constructed by curating a total of 781 known inhibitors, which were used to develop binary classification models using random forests to distinguish between highly and weakly active inhibitors. The Matthews correlation coefficient of the consensus model reached 0.771 for the test set. A virtual screening procedure of 3.4 million molecules was conducted based on shape similarity and predictions from the consensus model to identify four hits and a favorable benzothiazole moiety. A molecular generation model was trained using molecules from both the curated dataset and the identified hits to generate potential inhibitors, which were subsequently predicted by the consensus model. Finally, eight compounds were selected and synthesized for biochemical testing, leading to the identification of compound <b>19</b>, which had a novel scaffold and acceptable potency: inhibition rates of 80.7% at a concentration of 100 nM and 39.5% at a concentration of 10 nM. Compound <b>19</b> could serve as a suitable starting point for further optimization efforts in medicinal chemistry.</p><p></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

Design and discovery of POLQ helicase domain inhibitors by virtual screening and machine learning

  • 期刊论文

  • 发表日期：

    2025年5月8日

查看全文
  • Wei Feng,
  • Lei Liu,
  • Lingjun Li,
  • Peng Du,
  • Zhichen Yuan,
  • Jing Yuan,
  • Changjiang Huang,
  • Zijian Qin

摘要

DNA polymerase theta (Polθ or POLQ) is an attractive target for treating BRCA-deficient cancers. In the present work, several computational approaches were employed for the design and discovery of novel POLQ helicase domain inhibitors. A dataset was constructed by curating a total of 781 known inhibitors, which were used to develop binary classification models using random forests to distinguish between highly and weakly active inhibitors. The Matthews correlation coefficient of the consensus model reached 0.771 for the test set. A virtual screening procedure of 3.4 million molecules was conducted based on shape similarity and predictions from the consensus model to identify four hits and a favorable benzothiazole moiety. A molecular generation model was trained using molecules from both the curated dataset and the identified hits to generate potential inhibitors, which were subsequently predicted by the consensus model. Finally, eight compounds were selected and synthesized for biochemical testing, leading to the identification of compound 19, which had a novel scaffold and acceptable potency: inhibition rates of 80.7% at a concentration of 100 nM and 39.5% at a concentration of 10 nM. Compound 19 could serve as a suitable starting point for further optimization efforts in medicinal chemistry.