In a groundbreaking effort, a team of scientists has introduced a flagship AI dataset from a study investigating biomarkers and environmental factors potentially influencing the development of type 2 diabetes.
The study’s participants include both those without diabetes and individuals at various stages of the disease, and early findings suggest a complex array of insights distinct from prior research, according to a report published in Nature Metabolism.
“We’re seeing data that supports heterogeneity among type 2 diabetes patients—meaning that people aren’t all facing the same condition. Thanks to the large, detailed datasets we’re gathering, researchers will be able to examine this diversity more closely,” said Dr. Cecilia Lee, professor of ophthalmology at the University of Washington School of Medicine, US.
For instance, data from custom environmental sensors placed in participants’ homes reveal a clear association between disease progression and exposure to fine particulate pollutants.
The data also encompass survey responses, depression scales, eye-imaging scans, and traditional measurements of glucose and other biological indicators.
“All these data are intended to be mined by AI for new insights on risks, preventive strategies, and pathways linking disease and health,” the authors noted.
The study aims to gather health data from a racially and ethnically diverse population, surpassing prior datasets in representation, while ensuring the data is technically and ethically ready for AI analysis.
“This discovery process has been invigorating. We are a consortium of seven institutions with multidisciplinary teams who haven’t worked together before but share the goals of using unbiased data and safeguarding data security as we make it accessible to researchers worldwide,” said Dr. Aaron Lee, a UW Medicine professor of ophthalmology and the project’s principal investigator.
Hosted on a customized online platform, the dataset is available in two formats: a controlled-access version requiring a usage agreement and a publicly accessible version without HIPAA-protected information.
(Inputs from IANS)
