A Method for Extracting Pharmacokinetics Properties from Package Inserts: Usage of Interactive Artificial Intelligence Systems
- VernacularTitle:添付文書から薬物動態特性を抽出する方法の検討：対話型 AI システムの活用
- Author:
Tsuyoshi ESAKI
1
;
Keiko OGAWA
2
;
Kazuyoshi IKEDA
3
Author Information
- Keywords: data collection; large languages model; generative pre-trained transformer; pharmacokinetics
- From:Japanese Journal of Drug Informatics 2024;26(2):80-91
- CountryJapan
- Language:Japanese
- Abstract: Objective: Research and development for drug discovery is time-consuming and expensive. Artificial intelligence (AI) technologies, such as machine learning are attracting the attention of researchers as tools for efficiently advancing drug discovery. However, the use of AI technology requires a high amount of data, and the scope of application and accuracy of prediction depend on data quality. Therefore, the development of technology for efficiently collecting drug information data is required. The present examined an interactive AI system for extracting absorption, distribution, metabolism, and excretion (ADME) data from clinical practice documents. Methods: Attachments for five drugs were collected from the Pharmaceuticals and Medical Devices Agency (PMDA) for properties influencing pharmacokinetics, including dosage, maximum concentration (Cmax), half-life (T1/2), time to peak drug concentration (Tmax), area under the curve (AUC), and clearance (CL). Data were collected directly from PDFs using ChatGPT Plus, SciSpace, and ChatPDF as interactive AI systems capable of performing this task, and variations in these properties were compared. In addition, we compared the variations in the prompting outputs. Results: ChatGPT Plus was able to retrieve some pharmacokinetic properties including the values in the tables, whereas SciSpace and ChatPDF were unable to retrieve pharmacokinetic information. In addition, the ChatGPT Plus output changed depending on the prompt, whereas the results obtained using SciSpace and ChatPDF did not change significantly based on the prompt. Therefore, ChatGPT Plus was the most appropriate system for collecting ADME data. Conclusion: Based on the results of collection of ADME characteristics from documents using the three interactive AI systems, ChatGPT Plus is the most effective method for obtaining the desired characteristics, although several issues need to be addressed. Interactive AI will be an indispensable technology for data collection in drug research, and could contribute significantly to drug discovery in the future.
