Scholarlink

<p>We present <i>SangsterLogP</i>, the largest publicly available curated dataset of experimental <i>logP</i> values, comprising more than 23k unique molecules, with experimental <i>logP</i> values ranging from −3.8 to 11.7 (about 15.9 log units). The dataset originated from Dr. James Sangster’s comprehensive literature review of over 3k sources. We implemented a systematic curation workflow including a) <i>logD</i>-to-<i>logP</i> adjustment for ionised compounds and b) consensus-based residual analysis for outliers and duplicates removal. External validation using retrospective and prospective test sets demonstrated robust predictive performance (RMSE of 0.34 and 0.47 log units, respectively). <i>SangsterLogP</i> also substantially expands coverage of chemical space compared to the widely used legacy PHYSPROP database, including compounds in the beyond-Rule-of-5 domain. The fully annotated dataset, including experimental conditions and sources, is freely accessible via the Zenodo repository and on the Online Chemical database and Modelling Environment website</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

SangsterLogP - the largest publicly available dataset of logP values

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月7日

查看全文
  • Thalita Cirino,
  • Giulia Caron,
  • Giuseppe Ermondi,
  • Larysa Charochkina,
  • Igor V. Tetko

摘要

We present SangsterLogP, the largest publicly available curated dataset of experimental logP values, comprising more than 23k unique molecules, with experimental logP values ranging from −3.8 to 11.7 (about 15.9 log units). The dataset originated from Dr. James Sangster’s comprehensive literature review of over 3k sources. We implemented a systematic curation workflow including a) logD-to-logP adjustment for ionised compounds and b) consensus-based residual analysis for outliers and duplicates removal. External validation using retrospective and prospective test sets demonstrated robust predictive performance (RMSE of 0.34 and 0.47 log units, respectively). SangsterLogP also substantially expands coverage of chemical space compared to the widely used legacy PHYSPROP database, including compounds in the beyond-Rule-of-5 domain. The fully annotated dataset, including experimental conditions and sources, is freely accessible via the Zenodo repository and on the Online Chemical database and Modelling Environment website