Scholarlink

<p>The conventional separation of drug-target affinity (DTA) prediction and de novo molecule generation creates a significant bottleneck in drug discovery. To address this, we introduce MutiDTAGen, a unified multi-task learning framework that establishes a bidirectional system between these two complementary tasks. By utilizing shared deep representations and a dynamic optimization strategy, the proposed framework ensures that knowledge from affinity prediction directly guides the generation of target-specific molecules. Our method demonstrates improved performance across multiple benchmarks, achieving, for instance, a 12% reduction in Mean Squared Error (MSE) on the Davis dataset compared to the GraphDTA baseline. This synergistic approach not only enhances prediction accuracy but also improves the quality and target-specificity of generated compounds. By unifying prediction and generation within a single end-to-end architecture, this study offers a unified and efficient computational strategy for drug discovery.</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

MutiDTAGen: fusion framework of perceptual new drug generation and drug-target affinity prediction through multi-scale feature extraction

  • 期刊论文

  • 发表日期：

    2026年1月10日

查看全文
  • Xingyu Liu,
  • Xiaorui Huang,
  • Jirui Zhang,
  • Maoyuan Zhou,
  • Jiaxing Li,
  • Zhiwei Zhang,
  • Tianhao Liu,
  • Zhenghui Wang,
  • Nasrollah Moghadam,
  • Hossein Ganjidoust,
  • Qianjin Guo

摘要

The conventional separation of drug-target affinity (DTA) prediction and de novo molecule generation creates a significant bottleneck in drug discovery. To address this, we introduce MutiDTAGen, a unified multi-task learning framework that establishes a bidirectional system between these two complementary tasks. By utilizing shared deep representations and a dynamic optimization strategy, the proposed framework ensures that knowledge from affinity prediction directly guides the generation of target-specific molecules. Our method demonstrates improved performance across multiple benchmarks, achieving, for instance, a 12% reduction in Mean Squared Error (MSE) on the Davis dataset compared to the GraphDTA baseline. This synergistic approach not only enhances prediction accuracy but also improves the quality and target-specificity of generated compounds. By unifying prediction and generation within a single end-to-end architecture, this study offers a unified and efficient computational strategy for drug discovery.