Toward a Machine Learning-Driven Digital Twin for Real-Time Hormone Biosensing in Personalized Infertility Care

Area of Science:

• Biomedical Engineering
• Digital Health
• Precision Medicine

Background:

• Personalized healthcare demands advanced solutions beyond one-size-fits-all approaches.
• Digital twin (DT) technology offers real-time virtual replicas for personalized medicine via monitoring, simulation, and prediction.

Purpose of the Study:

• To develop a machine-learning-driven biosensor digital twin for personalized infertility treatment.
• To integrate the biosensor DT with a Smart Health Monitoring System.
• To replicate the behavior of a field-effect transistor (FET)-based biosensor using experimental data.

Main Methods:

• Trained a digital twin using experimental data from silicon nanonet BioFET prototypes functionalized with 17β-estradiol aptamers.
• Evaluated seven supervised machine learning algorithms to predict hormone concentration from electrical parameters (Vg, Isd).
• Utilized Leave-One-Biosensor-Out validation to assess cross-device generalization.

Main Results:

• The K-Nearest Neighbors (KNN) model demonstrated the highest predictive accuracy (R² = 0.99, CV-R² = 0.98, RMSE = 11.87 pg/mL).
• The KNN model showed robust cross-device generalization (R² = 0.59), confirming its ability to capture nonlinear relationships.
• The model successfully generalized across independently fabricated sensors.

Conclusions:

• The developed model serves as a validated predictive core for a biosensor digital twin.
• This framework supports data-driven biosensor digital twins for smart health monitoring in personalized infertility care.
• Future work will focus on real-time synchronization and closed-loop feedback for adaptive clinical applications.