PMx-AI Bot: Changing the way of traditional Pharmacometrics work with AI Bots
Ercan Suekuer
Roche
Objectives:
The burgeoning influence of artificial intelligence (AI) is revolutionizing the field of pharmacometrics. This transformation is not only due to the integration of machine learning (ML) algorithms, which have significantly improved the accuracy and efficiency of pharmacokinetic/pharmacodynamic (PK/PD) modeling, but also to the advent of generative AI (GenAI) technologies. Our innovative AI ChatBot, named PMxAIBot, is designed to be a real-time assistant for pharmacometricians, facilitating a range of tasks from NONMEM code generation to model suggestion and information synthesis from vast textual datasets.
Methods: The PMxAIBot is a cutting-edge tool based on the gpt4-turbo architecture from Microsoft Azure OpenAI. It has been meticulously "trained" using a comprehensive compilation of datasets, including proprietary data from Roche, publicly available pharmacometric research, clinical trial data, PK/PD models, and insightful publications from key competitors in the pharmaceutical industry. To rigorously evaluate the bot's performance, we subjected it to a series of standard pharmacometric tasks, including model selection, detailed covariate analysis, and complex simulation studies, all of which are pivotal in the drug development process.
Results: Incorporating PMxAIBot into the pharmacometric workflow has led to a remarkable increase in productivity among users. Routine tasks, such as programming and model implementation, are accomplished more rapidly with the bot's assistance. Furthermore, PMxAIBot has demonstrated exceptional capability in guiding users to identify optimal models for their datasets, suggest pertinent covariates, and generate simulation outputs that are instrumental in the decision-making process. The AI bot has exhibited a high concordance rate with the choices and recommendations made by seasoned pharmacometricians. Feedback from users has consistently highlighted the bot's user-friendly interface, the remarkable speed with which it conducts analyses, and the high caliber of insights produced by the AI.
Conclusions: The development of PMxAIBot marks a significant leap forward in the pharmacometric arena, positioning it as a powerful AI collaborator that augments the expertise of pharmacometricians. By harnessing the analytical prowess of gpt-4 Turbo, PMxAIBot has not only simplified the model development workflow but has also provided invaluable educational assistance, thereby accelerating the timeline of drug development initiatives. While the bot's current iteration shows great promise, it is not without its limitations; further refinement is necessary, particularly in the preprocessing of input documents, to fully optimize its performance. The efficacy of PMxAIBot is also dependent on the user's proficiency in communicating with the AI, which involves the art of prompt engineering and posing the right questions. This skill set is critical and represents a significant learning curve for pharmacometricians. As users become more proficient in interacting with the bot, we anticipate an even more pronounced enhancement in productivity and the quality of decision-making. PMxAIBot exemplifies the transformative potential of AI within the domain of pharmacometrics and sets the stage for continued innovation in the field. It also underscores the importance of ongoing education and adaptation in the era of AI-assisted drug development, highlighting the intersection of technology and human expertise as a driver for progress in the pharmaceutical industry.