Identification and Quantification of Multiple Drugs by Machine Learning on Electrochemical Sensors for Therapeutic Drug Monitoring

Electrochemical sensors play a pivotal role in advancing therapeutic drug monitoring (TDM) and personalized therapy. However, the insufficient selectivity of electrochemical sensors based on cytochromes P450 poses a great need for innovative machine learning (ML) integrated with cyclic voltammograms obtained from electrochemical sensors and cyclic voltammetry to construct a fully automated system for accomplishing the precise identification and quantification of multiple drugs in the patients' blood. An effective ML-based method is proposed in this letter with the adaptive transformation of Gaussian mixture model algorithm for Gaussian feature extraction from the redox peak with peculiar characterisation and a novel artificial neural network-based model IdentQuantNet for the drug identification and quantification. The simulated dataset and measured dataset of cyclophosphamide (CP) and ifosfamide (IFO), proposed here as model drugs, are demonstrated to validate the efficacy of our proposed method. Our method can achieve the precision at 100% for identification task and the mean absolute error at 4.88 and 6.67 for CP and IFO for the quantification task on the measured dataset.