通过错误纠正提示工程改进药物不良反应的大型语言模型
1School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China.
Journal of biomedical informatics
|August 30, 2025
概括
本研究使用大型语言模型引入了一种用于药物不良反应 (ADR) 识别的新型提示模板. 这种新方法显著提高了识别药物名称和不良反应的准确性,提高了患者的安全性.
科学领域:
- 药物监督管理
- 自然语言处理
- 医学的人工智能
背景情况:
- 药物不良反应 (ADRs) 的监测对于患者的安全性和治疗疗效至关重要.
- 通过命名实体识别 (NER) 准确识别药物名称,成分和不良反应对于药物安全和信息整合至关重要.
- 由于数据的可变性和药物名称的相似性,现有的NER方法难以获得ADR,通常需要大量的手动注释.
研究的目的:
- 使用大型语言模型 (LLM) 提出一个有效的ATS实体识别提示模板.
- 在复杂的医学文本中提高药品名称,成分和不良反应的准确性.
- 提供一个强大的方法来提取与毒品相关的实体和构建知识图表.
主要方法:
- 开发了一个ADR提示模板,包含任务描述,实体解释,指导方针,少量学习样本和错误纠正示例.
- 整合了来自网络的复杂的ADR数据,并使用Begin,Inside,Outside (BIO) 标注方法创建了一个集体.
- 用GPT-3.5和GPT-4评估提示模板的有效性,将结果与微调的LLaMA和DeepSeek模型进行比较.
主要成果:
- 提议的提示模板显著提高了GPT-3.5的F1分数,从0.648提高到0.887.
- 随着新的提示模板,GPT-4的F1分数从0.757增加到0.921.
- 与微调的LLaMA和DeepSeek模型相比,该方法显示出更高的性能.
结论:
- 新的提示模板大大提高了LLM中的ADR实体识别准确性.
- 这种方法提供了对现有的NER方法进行ADR分析的优质替代方案.
- 这些发现为未来的毒品相关实体关系提取和知识图构建提供了坚实的基础.
相关概念视频
Pharmacovigilance
983
Post-marketing surveillance is a critical component of pharmaceutical regulation, often uncovering unanticipated adverse drug reactions (ADRs) once a drug is widely used over an extended period.
This process, termed pharmacovigilance, aims to detect, evaluate, and minimize harmful effects related to medication use. The data collection for pharmacovigilance depends on spontaneous reporting systems, where healthcare professionals or patients voluntarily report suspected ADRs.
In some cases, there...
This process, termed pharmacovigilance, aims to detect, evaluate, and minimize harmful effects related to medication use. The data collection for pharmacovigilance depends on spontaneous reporting systems, where healthcare professionals or patients voluntarily report suspected ADRs.
In some cases, there...
983
Drug Nomenclature
2.3K
During the development of a new pharmaceutical, the manufacturer initially assigns a code name to the drug. Once approved, the drug receives a United States Adopted Name (USAN)—a generic, nonproprietary designation. Upon being listed in the United States Pharmacopeia, this nonproprietary name becomes the drug's official name. Additionally, the manufacturer assigns a proprietary name or trademark, which serves as the brand name under which the drug is marketed. It is worth noting that...
2.3K
Improving Translational Accuracy
2.7K
2.7K
Allergic Drug Reactions
972
Allergic reactions related to drugs are hypersensitivity responses driven by the immune system and bear no connection to the drug's therapeutic action. While drugs in isolation do not trigger an immune response, they can interact with endogenous proteins to form antigens. These antigens stimulate lymphocytes to produce antibodies. IgE-type antibodies attach themselves to mast cells. Upon subsequent exposure to the same stimulus, the antigen-antibody interaction is initiated, unleashing...
972
Anticholinesterase Agents: Poisoning and Treatment
1.0K
Anticholinesterases, also known as cholinesterase inhibitors, work by blocking the breakdown of acetylcholine, leading to its accumulation in the synaptic cleft. This accumulation indirectly enhances both muscarinic and nicotinic actions. These agents are classified as reversible or irreversible based on their mechanism of action.
Irreversible agents form a strong bond with the cholinesterase enzyme, making it inactive. The breakdown of the phosphorylated enzyme is...
Irreversible agents form a strong bond with the cholinesterase enzyme, making it inactive. The breakdown of the phosphorylated enzyme is...
1.0K
Drug Metabolism: Phase II Reactions
4.1K
Phase II reactions are essential for the detoxification and elimination of drugs from the body. These reactions involve the conjugation of parent drugs or their phase I metabolites with endogenous molecules, resulting in more hydrophilic drug conjugates. The primary conjugation reactions in this phase are sulfation and glucuronidation. Both sulfation and glucuronidation typically produce biologically inactive metabolites. However, in some cases involving prodrugs, active metabolites may be...
4.1K


