Researchers Evaluate AI as a Tool for Generating New Research Hypotheses

Researchers from the Celedón Laboratory for Pediatric Asthma Research in the Division of Pediatric Pulmonary Medicine at UPMC Children's Hospital of Pittsburgh have developed and tested an artificial intelligence (AI)-assisted approach designed to help researchers identify promising new research questions from large clinical datasets. The proof-of-concept study, published in the Journal of Respiratory Biology and Translational Medicine, used childhood asthma to evaluate whether AI could make one of the most time-consuming steps in epidemiologic research more efficient.

About the Study and Possible Implications for the Research Community

Modern research databases contain vast amounts of clinical, laboratory, and demographic variables. Determining which observations are already supported by published evidence, which may represent meaningful new research questions to ask, and which are perhaps unimportant requires extensive statistical analysis and literature review, along with time and expertise. The Celedón Lab team developed an AI-assisted process designed to help streamline that process but leaving the scientific interpretation and decision-making to actual researchers who have the necessary background and skillsets.

To test this approach, the team analyzed data from more than 25,000 children participating in the National Health and Nutrition Examination Survey (NHANES). After identifying statistical associations with childhood asthma, the AI system searched the scientific literature and evaluated whether each finding was already supported by published evidence or represented a potential area for new exploration or consideration. Human investigators then reviewed the results.

In short, the framework successfully distinguished well-established asthma associations from findings with only indirect supporting evidence and those lacking meaningful evidence in the published literature.

AI tools and models, when properly used, controlled, and interpreted, can help organize and synthesize large quantities of existing scientific knowledge rapidly, allowing investigators to spend less time searching the literature and more time evaluating the biological significance of potential discoveries.

Although childhood asthma was the test case for the model, this same approach could be applied to many other diseases that rely on large observational datasets. Instead of replacing researchers, these kinds of AI and LLM tools may help accelerate hypothesis generation while keeping in place the scientific standards upon which all research must necessarily rest.

Read more about the team’s methodology, analysis, and interpretations using the study link below. You can also read more about the Celedón Laboratory for Pediatric Asthma Research, its team members, areas of focus, and other recent studies by visiting the lab’s website.

Reference

Liu J, Han Y-Y, Ye X, Rosser FJ, Gaietto KM, Zhao C, Chen W, Celedón JC. Harnessing Artificial Intelligence for Hypothesis Generation in Childhood Asthma: Insights From NHANES. J Respir Biol Transl Med. 2026; 3(2): 10003.