Scholarlink

<p>Artificial intelligence is transforming drug discovery by enabling exploration of vast chemical spaces and the design of new molecules with tailored properties. However, most approaches focus on chemical features or protein binding, leaving phenotypic discovery underexplored. Here, we present a framework to design small molecules with selective toxicity in pancreatic cancer cells by combining large-scale screening data with AI models. Using data from over 11,000 compounds tested across cancer and control cell lines, we trained predictive models and integrated them into a generative system to design new candidates. This approach produced novel molecules with desired cell-specific effects, many of which were structurally distinct from known compounds. Experimental validation confirmed that several designed molecules showed the intended selective activity, outperforming conventional screening strategies. Overall, this work highlights the potential of combining predictive and generative AI to design compounds with complex biological effects without relying on predefined targets.</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

Phenotypic AI-based design of cell-specific small molecule cytotoxics

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月20日

查看全文
  • Gema Rojas-Granado,
  • Marta Sánchez-Soto,
  • Jesús Calahorra,
  • María Caballero,
  • Israel Ramos,
  • Martino Bertoni,
  • Patrick Aloy

摘要

Artificial intelligence is transforming drug discovery by enabling exploration of vast chemical spaces and the design of new molecules with tailored properties. However, most approaches focus on chemical features or protein binding, leaving phenotypic discovery underexplored. Here, we present a framework to design small molecules with selective toxicity in pancreatic cancer cells by combining large-scale screening data with AI models. Using data from over 11,000 compounds tested across cancer and control cell lines, we trained predictive models and integrated them into a generative system to design new candidates. This approach produced novel molecules with desired cell-specific effects, many of which were structurally distinct from known compounds. Experimental validation confirmed that several designed molecules showed the intended selective activity, outperforming conventional screening strategies. Overall, this work highlights the potential of combining predictive and generative AI to design compounds with complex biological effects without relying on predefined targets.