标题： 通过与大型语言模型合作，推动基于结构的药物设计的边界 显示英文标题

标题： Pushing the boundaries of Structure-Based Drug Design through Collaboration with Large Language Models

摘要： 基于结构的药物设计（SBDD）通过使分子针对特定蛋白质靶点的合理设计，彻底改变了药物发现。 尽管在提高对接分数方面取得了显著进展，先进的3D-SBDD生成模型在产生符合药物化学标准和药代动力学要求的药物样候选物方面仍然面临挑战。 这些限制源于其对分子相互作用的固有关注，常常忽视了药物样性的关键方面。 为了解决这些不足，我们引入了协作智能药物设计（CIDD）框架，该框架结合了3D-SBDD模型的结构精度与大型语言模型（LLMs）的化学推理能力。 CIDD首先通过3D-SBDD模型生成支持分子，然后通过LLM支持的模块对这些分子进行优化，以增强药物样性和结构合理性。 在CrossDocked2020数据集上的评估中，CIDD实现了37.94%的显著成功比例，明显优于之前的最先进基准15.72%。 尽管改善分子相互作用和药物样性通常被视为一种权衡，但CIDD通过利用不同模型的互补优势，独特地实现了两者的平衡改进，为设计具有治疗潜力的药物候选物提供了一条稳健且创新的途径。 显示英文摘要

摘要： Structure-Based Drug Design (SBDD) has revolutionized drug discovery by enabling the rational design of molecules for specific protein targets. Despite significant advancements in improving docking scores, advanced 3D-SBDD generative models still face challenges in producing drug-like candidates that meet medicinal chemistry standards and pharmacokinetic requirements. These limitations arise from their inherent focus on molecular interactions, often neglecting critical aspects of drug-likeness. To address these shortcomings, we introduce the Collaborative Intelligence Drug Design (CIDD) framework, which combines the structural precision of 3D-SBDD models with the chemical reasoning capabilities of large language models (LLMs). CIDD begins by generating supporting molecules with 3D-SBDD models and then refines these molecules through LLM-supported modules to enhance drug-likeness and structural reasonability. When evaluated on the CrossDocked2020 dataset, CIDD achieved a remarkable success ratio of 37.94%, significantly outperforming the previous state-of-the-art benchmark of 15.72%. Although improving molecular interactions and drug-likeness is often seen as a trade-off, CIDD uniquely achieves a balanced improvement in both by leveraging the complementary strengths of different models, offering a robust and innovative pathway for designing therapeutically promising drug candidates.