(333h) Machine Learning-Enabled Electrochemical Gas Sensing of Volatile Organic Compounds Using a Microfluidic Membrane Contactor

Volatile Organic Compounds (VOCs) are a class of organic compounds that exhibit high vapor pressure at ambient conditions, resulting in their predominant presence in the gaseous phase at room temperature. Recently, VOCs have gained interest in biochemical and medical diagnostics. Elevated levels of certain VOCs in human breath have been identified as potential biomarkers for various cancers, offering a non-invasive approach to early cancer detection. Decades of research have found that plants emit different VOCs, which are correlated to multiple biotic and abiotic stresses in plants. The current gold-standard approach for identifying VOCs with high sensitivity and selectivity involves gathering the sample using collection cartridges or containers, followed by laboratory-based analysis utilizing Gas Chromatography-Mass spectrometry. Nevertheless, this approach necessitates sample preparation, transportation, and the desorption of gases, all contributing to errors in analysis, substantial turnaround times, and augmenting the overall expenses of the procedure. Considering these limitations, our research focused on developing a portable, low-cost, highly sensitive, and selective VOC sensor for point-of-use (POU) and point-of-care (POC) applications. Most research focuses on electronic noses using metal oxide semiconductor (MOS) arrays to measure conductivity changes in response to VOCs. However, MOS sensors are known for their poor selectivity. In contrast, our electrochemical sensor leverages ionic liquids (ILs) as the sensing elements, providing significantly enhanced selectivity. This improvement is achieved by either tuning the ions or employing different ILs with varying gas solubilities. Additionally, our sensor uses multiple electrochemical techniques like electrochemical impedance spectroscopy (EIS), cyclic voltammetry, and amperometry to measure conductivity, charge transfer resistance, capacitance, and diffusion, offering richer data for AI/ML-based predictions, which can enhance sensitivity and selectivity.

To incorporate the ILs and perform in-situ detection, we developed a sensor architecture integrating microfluidics and microelectrode technology by leveraging advanced, scalable microfabrication techniques like photolithography and physical vapor deposition. The first layer features a glass slide with two interdigitated gold microelectrode arrays, each 100 nm thick. The second layer consists of double-sided tape with a microchannel to confine the IL. This microchannel is precisely aligned to position the microelectrodes within it, allowing for direct contact with the IL once it is introduced. The third layer includes a hydrophobic Polytetrafluoroethylene (PTFE) membrane with polypropylene support, with the smooth PTFE side facing the liquid microchannel. The fourth layer comprises double-sided tape with a microchannel for continuous gas flow. Finally, the fifth layer consists of another glass slide, which features fluid ports for connecting tubing to facilitate gas flow. This multi-layered design, which is essentially a membrane contactor on a microscale, enables continuous gas flow through the device, allowing interaction between the gas and the IL via the membrane, and facilitates the detection of perturbations in the IL using microelectrodes connected to a potentiostat.

Initial experiments were performed to understand the baseline stability of the electrochemical sensor integrated with ILs using EIS. A temporal drift in the impedance spectrum was observed in this study. A detailed investigation into this provided insight into the fact that the drift was due to the interaction of IL with humidity in the membrane pores and gas microchannel. This unreliable baseline drift in the electrochemical response was mitigated by using hydrophobic ILs, as they show minimum interaction with humidity. Consequently, hydrophobic ILs were used to detect model VOCs acetone and toluene. Commercially available ILs 1-Ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide ([EMIM][TFSI]) and 1-Butyl-3-methylimidazolium hexafluorophosphate ([BMIM][PF₆]) were integrated into the microfluidic platform for selectively detecting acetone and toluene. Different electrochemical techniques like EIS, cyclic voltammetry, differential pulse voltammetry, and amperometry were employed to detect the perturbations caused by acetone and toluene in both ILs. The data obtained from all these techniques were used to train the support vector machine ML model without the use of any electrochemical models (or equivalent circuit models commonly used in EIS) and distinguish the signals corresponding to acetone and toluene. Further, the support vector regression ML model was used to create calibrations for the sensor.