(P08-019-25) AI ChatGPT Application for Diet Validity in Rodent Research

Objectives: Application of Artificial Intelligence (AI) using a large language model like OpenAI’s ChatGPT is a rapid means to parse rodent articles that used diets. A generative AI model, guided by targeted prompts, can analyze and catalog rodent research by obtaining nutrient compositions, diet types, research objectives, and outcomes to facilitate future research and peer review.

Methods: ChatGPT was used to interrogate published articles that used diets in rodent research (formulations, ingredients, and nutrients). Prompts identified experimental designs, methods, objectives, statistics, and findings for the research hypothesis. The model used was gpt-4o with rigid constraints using ‘chow diet’, purified diet’, or ‘semi-purified diet’ to parse results in 2 phases. In the tuning phase, inference with ChatGPT models was done half well-defined studies and half less well-defined studies. In this phase, prompts were iteratively developed by evaluating ChatGPT inference results after each run. The testing phase ascertained if our results could scale to a larger sample of articles, which has the same occurrence of well-defined and less well-defined studies.

Results: Our study confirms that ChatGPT can recover diet and ingredient information from rodent studies to validate experimental designs aimed at specific research objectives. The testing of prompts can be used to create rubrics for articles using diets in rodent studies. We found that subjecting articles to ChatGPT models via the API the files must be converted from pdf to raw text. The various ways a pdf can have formatted text presents challenges during this process which can yield poor results from GenAI models.

Conclusions: ChatGPT can evaluate rodent diets and nutrients to assess study designs, research objectives, and validity for confirming findings. Applications of ChatGPT data can be used in peer review or in developing appropriate experimental designs for future nutrition research. Our study results fill the need for rapid and comprehensive literature searches of diets to improve experimental designs and information gathering for scientists performing nutrition and biomedical research with rodents.

Funding Sources: No extramural funding was used.