Nanoelectronic Detection of Opioids: Machine Learning-Powered Screening With Carbon Nanotube Field-Effect Transistor Sensor Array.

The development of inexpensive, portable, and rapid opioid screening techniques is urgently needed because of the ongoing opioid crisis in the United States, primarily driven by illicit fentanyl. In this work, a nanoelectronic detection method was developed for opioids using machine learning. Fentanyl and three other opioids (codeine, hydrocodone, and morphine) were tested with field-effect transistor (FET) sensor array composed of single-walled carbon nanotubes (SWCNTs) decorated with gold and platinum nanoparticles. Hundreds of opioid samples were tested manually and with an automated electrolyte-gated FET test system capable of testing up to 96 sensors. Fifteen features were extracted from experimental FET data to train conventional supervised machine learning models. Optimized linear discriminant analysis (LDA) and random forest (RF) models were applied for opioid classification, achieving 82.6% accuracy. Deep learning methods were subsequently applied to the full FET dataset to eliminate human bias in feature selection. A convolutional neural network (CNN) achieved 98.3% accuracy for the manual test group and revealed the significant role of gate-source current in distinguishing opioid molecules. This insight guided improvements to conventional models, uncovering unexpected mechanisms of molecular sensing with liquid-gated FETs and demonstrating the broader potential of combining nanoelectronics with machine learning for chemical detection.