Quantitative Estimation of Sensitivity of Lipolysis to Insulin

National Institute Of Diabetes And Digestive And Kidney Diseases

Investigators

Linked publications, trials & patents

Abstract

Obesity and insulin resistance remain major public health challenges, with elevated plasma free fatty acids (FFA) playing a central role in impairing insulin action across tissues and altering downstream signaling pathways. Quantitative evaluation of insulin’s regulation of glucose and FFA following intravenous glucose challenge is therefore critical for assessing insulin resistance in both research and clinical settings. Mathematical minimal models (MMs) based on insulin-modified frequently sampled intravenous glucose tolerance tests (IM-FSIGT) have been widely used to derive an insulin sensitivity index, and our previous work extended these models to include FFA dynamics. While this extension provided a useful index for the sensitivity of lipolysis to insulin, the minimal structure limited the ability to capture FFA behavior at later times after the insulin bolus. Moreover, conventional parameter optimization methods for nonlinear ODE models are slow, initialization-dependent, and difficult to constrain to physiological ranges and covariances, creating barriers to reproducible, physiologically valid inference. We developed and validated a deep-learning framework that overcomes these limitations by replacing iterative optimization with one-shot parameter inference. Using simulated datasets generated from physiologically plausible parameter distributions, we trained convolutional neural networks (CNNs) to map FSIGT time courses of glucose, insulin, and FFA directly to underlying model parameters. The trained networks consistently recovered parameters obtained by traditional optimization, accurately reconstructed model-fit trajectories, and generalized robustly to experimental FSIGT data. We systematically evaluated the effect of training dataset scale, feature engineering, and activation function choices, showing that larger datasets and appropriate feature engineering improve accuracy and identifiability. This framework delivers fast, physiologically consistent parameter estimates that enable robust fitting of glucose–insulin–FFA dynamics and establish the basis for a quantitative index of insulin’s acute regulation of lipolysis. Looking ahead, a key challenge is extending the FFA minimal model to capture later phases of FFA dynamics observed in long-duration (>4 h) FSIGT protocols, which remains an active objective for our research.