Decoding Type 2 Diabetes Progression via Metabolic Hormone Time-Series

Unraveling the dynamics of diabetes hormones and peptides, including insulin, glucagon, and C-peptide, is important for understanding diabetes progression and for personalized treatment. The current gold standard methods to monitor these biomarkers have a prolonged processing time, and their protocols do not allow real-time data collection. We recently developed the quantum dot–integrated real-time ELISA (QIRT-ELISA), a continuous monitoring platform that quantifies these key biomarkers at minute-by-minute resolution. Here, we leverage QIRT-ELISA’s high temporal resolution for real-time, multiplexed monitoring of insulin, glucagon, and C-peptide in animal models of prediabetes and type 2 diabetes, generating the first in vivo, hormone time-series dataset. Using the raw QIRT-ELISA time series, we trained and evaluated four classical machine-learning classifiers to classify metabolic phenotypes, a capability not achievable with conventional assays or with unprocessed QIRT-ELISA outputs. Among these classifiers, logistic regression performed the best, achieving 89% accuracy in distinguishing healthy, prediabetic, and diabetic states. QIRT-ELISA’s capacity to capture high-resolution, time-series hormone data, along with machine-learning–based classification, has the potential to deconvolute the roles of these hormones in the onset and progression of diabetes.