Optimized IdentQuantNet: A machine learning-based approach for identification and quantification of multiple drugs with interaction on electrochemical sensors in personalized medicine

Electrochemical sensors are vital for advancing therapeutic drug monitoring (TDM) and personalized medicine by enabling precise identification and quantification of multiple drugs. However, fully automated machine learning (ML)-based models for simultaneously identifying and quantifying interacting drugs remain underdeveloped. To achieve better performance, such models require tailored loss functions and architectures that enhance accuracy when analyzing raw cyclic voltammogram (CV) redox peaks. This paper introduces Optimized IdentQuantNet, an improved ML-based framework for accurate identification and quantification of both interacting drugs and single agents, which combines a Preprocessor and ResultCombiner to analyze CVs obtained from electrochemical sensors. The framework extracts Gaussian features from redox peaks in single CVs to serve as input for the proposed model, facilitating drug identification and quantification, while utilizing a dual-loss approach to balance precision with clinical priorities, such as reducing underestimation. A multilayer perceptron (MLP)-based identification network (IdentNet) ensures reliable drug identification, while a quantification network (QuantNet) utilizes multi-branch and interaction-aware designs to handle interacting drug pairs and tailor single-agent quantification. Validation on simulated and measured datasets from etoposide (ETO), methotrexate (MTX), ifosfamide (IFO), cyclophosphamide (CP), and 5-Fluorouracil (5-FU) demonstrates the robustness and versatility of Optimized IdentQuantNet, highlighting its potential as a powerful tool for multi-drug electrochemical analysis and the progression of personalized medicine.