Biomimic models for in vitro glycemic index: Scope of sensor integration and artificial intelligence

Abstract

The accurate quantification of glycemic index (GI) remains crucial for diabetes management, yet current methodologies are constrained by resource intensiveness and methodological limitations. In vitro digestion models face challenges in replicating the dynamic conditions of the human gastrointestinal tract, such as enzyme variability and multi-time point analysis, leading to suboptimal predictive accuracy. This review proposes an integrated technological framework combining non-enzymatic electrochemical sensing with artificial intelligence to revolutionize GI assessment. Non-enzymatic sensors offer superior stability and repeatability in complex matrices, enabling real-time glucose quantification across multiple timepoints without enzyme degradation constraints. Machine learning algorithms, both supervised and unsupervised, enhance predictive accuracy by elucidating complex relationships within digestion data. This technological convergence represents a paradigm shift in food science analytics, promising improved throughput and precision in GI assessment. Future developments should focus on system scalability and broader applications across nutritional science, advancing diabetic management and personalized nutrition strategies.

Keywords: Artificial intelligence; Electrochemical sensor; Glycemic index; In vitro models; Starch hydrolysis.

Graphical abstract

Fig. 1

General scheme of carbohydrate digestion and glycemic response.(a) Oro-gastro-intestinal model. Carbohydrate digestion begins in the oral cavity, progresses through the gastrointestinal tract, and the majority of hydrolysis takes place in the small intestine. (b) Oral phase. Salivary α-amylase hydrolyzes starch (amylose and amylopectin) into α-limit dextrins, maltose, maltotriose, and various oligosaccharides, but cannot cleave α-1,6 bonds of amylopectin. The bolus formed is propelled into the stomach. (c) Gastric phase. Salivary α-amylase remains active briefly until gastric pH falls below 2. Minimal carbohydrate digestion occurs, but gastric processes disrupt carbohydrate-protein complexes. Chyme moves to the intestine via gastric emptying. (d) Intestinal phase and glucose absorption. Cholecystokinin (CCK) from acinar cell trigger the release of pancreatic α-amylase and hydrolyzes starch into smaller dextrins and maltose. Brush border enzymes (lactase, sucrase-isomaltase, glucoamylase) complete hydrolysis into monosaccharides. Lactase hydrolyzes lactose into glucose and galactose, while the sucrase-isomaltase complex splits sucrose into glucose and fructose and cleaves α-1,6 bonds in α-limit dextrins. Glucoamylase breaks down oligosaccharides (G2-G9) into glucose units. Glucose is absorbed by SGLT1 and transported into the bloodstream via GLUT2, leading to postprandial hyperglycemia. (e) Glycemic response and GI estimation. Blood glucose changes post-ingestion vary by GI: high GI (rapid spike), medium GI (moderate rise), and low GI (lower spike). GI is calculated by comparing the iAUC of test and reference foods. (iAUC)_T & (iAUC)_R are the incremental area under the curve of test food and reference food respectively.

Fig. 2

Proposed workflow for a non-enzymatic electrochemical glucose nanosensor with machine learning algorithms for estimating glycemic response(a) Optimization of in vitro GI estimation models: An in vitro method is optimized using a well-characterized food variety with a validated in vivo GI, subjected to simulated digestion phases. The method with the best correlation to in vivo GI is selected for high-throughput GI screening. (b) Development of electrochemical glucose nanosensor. The sensor is fabricated with functionalized electrodes and characterized using SEM, TEM, EDS, EIS. It is optimized for glucose detection with cyclic voltammetry and amperometry across a wide concentration range and evaluated for selectivity against interfering compounds. (c) Data integration and machine learning. Sensor and laboratory data are processed and analyzed using machine learning algorithms to develop predictive models, which are refined for optimal performance. (d) Integration with in vitro digestion of unknown samples. Food samples undergo in vitro digestion with optimized model, with glucose kinetics monitored via the electrochemical sensor. Sensor responses are compared to a calibration curve to estimate pGI using prediction model, validating the sensor's accuracy in reflecting glycemic response. (Scanning Electron Microscopy (SEM), Transmission Electron Microscopy (TEM), Electrochemical Impedance Spectroscopy (EIS), Energy-Dispersive X-ray Spectroscopy (EDS), SHX: Starch Hydrolysis at Specific Time Points e.g., (SH30 refers to starch hydrolysis at 30 min), AUC: Area Under the Curve, TS: Total Starch).