The workflow of our developed electro-olfactory platform is shown in Fig. 1 and consists of three steps: olfactory performance measurement, odor feature extraction, and odor TDI performance analysis. A key component of this platform is the sensor device, a graphene dispersion prepared by exfoliating graphite using 8-aminopyrene-1,3,6-trisulfonic acid trisodium salt (APTS) in aqueous solution. is built using Usual approaches for the functionalization of graphene sensors with pyrene derivatives rely on post-deposition step modification,33 Here, exfoliation and functionalization occur simultaneously.34 Therefore, APTS plays a dual role as a stabilizer to keep graphene dispersed and as an active adsorption site for analyte molecules.twenty four Microelectrodes are crosslinked to graphene flakes via dielectrophoresis (DEP) of drop-cast graphene dispersions under alternating current (AC), resulting in chemiresistor-type gas sensors. The odor sensing measurement setup consists of three modules: the odor steam generation module, the odor sensing measurement module, and the external signal display module. As shown in Fig. 1(a), we perform individual odor tests and binary odor mixture tests that mimic the sniffing stick test.
The raw data obtained from the odor sensing setup is the change in sensor current of a single graphene-based gas sensor as a function of operating time, with 24 replicates for each odor (Euca, Euge, 2Phe, and 2Nona). It consists of This iteration was performed to collect as much raw data as possible for machine learning in subsequent steps. In real-world applications, one iteration of detection analysis is sufficient once the machine learning classification model is established and deployed on an electronic olfactory platform. In this study, instead of steady-state features, 11 transient features (be1b1c1,2b2c2S, kmaximumkminutes,minutes, and area) extracted from the sensing response profile are subjected to machine learning analysis as shown in Fig. 1(b). exponential features (be1b1c1,2b2and c2) are the fitting parameters extracted from the sensing response profile and sensing recovery profile after exponential fitting, respectively.transient characteristics S. is the transient sensory response during the odor exposure phase, characterized by area Area under the sensory response curve during both odor exposure and odor flash periods.Temporary features kmaximum and kminutes represent the maxima and minima present in the sensing response profile after the first derivative fitting transformation. beminutes represents the minimum value of the transformed sensing response profile after the second derivative fitting transformation. Details of the feature extraction method are described in the Supplementary Material. Following feature extraction, both unsupervised and supervised machine learning analysis are performed to mimic the sniffing stick test.Finally, the olfactory performance of the developed electronic olfactory platform, i.e.,, Odor threshold performance, odor discrimination performance, and odor discrimination performance are evaluated as shown in Fig. 1(c).
