Multiplex Network-Based Approach to Gas Sensing

We present a multiplex network framework for analyzing sensor array data, integrating information from multiple network layers corresponding to different sensing conditions. Principal Component Analysis (PCA) is employed to reduce the dimensionality of the data and then used to construct the multiplex networks where each layer of the network is derived from correlation-based planar maximally filtered graphs (PMFG) for each distinct gas. Topological properties of the network layers are studied over different time frame. Clustering coefficients shows a persistent local connectivity for ethanol and ethylene (stabilizing near 0.71) but a decreasing activity from 0.70 to 0.56 for acetone over time. We observe a highest path length for the initial phase of the sensor lifecycle, and a general decrease in path length is observed for most gases over time, indicating a decreasing sensor sensitivity. The analysis of multidegree across different gases reveals both unique and shared connectivity patterns within sensor responses. Each gas shows some distinctive features in multi-degree which decreases over time, and increase in significant overlapping features between gas combinations. For example, a high multidegree (001110) indicates simultaneous unique features in sensor response for ammonia, acetaldehyde, and acetone and these features are not present in any other gas or their combination. We also observe a strong similarity in sensor response patterns between ethylene and ammonia. The shared features among certain gases indicate similar interaction patterns with sensors, potentially due to common chemical properties or response behaviors. We find that sensor drift significantly affects sensor responses, leading to evolving topological properties over time. Gas-specific network layers exhibit distinct trends, with some gases maintaining strong connectivity while others weaken due to drift. The methodology can be extended to other sensor arrays and applications beyond gas sensing, making it a versatile tool for multivariate sensor data analysis.