Convolutional Neural Networks on Correlation between GC−MS Molecular Data and QCM Gas-Sensing Data

Here, we demonstrate a correlation between gas chromatography/mass spectrometry (GC-MS) compositional data and quartz crystal microbalance (QCM) sensing data by developing a one-dimensional convolutional neural network (1D-CNN) model via principal component analysis (PCA). The 1D-CNN model was trained to predict the principal component scores derived from GC-MS profiles by using signals of nanostructured-QCM sensors with different surface modifications (ZnO, SnO₂, MgO, and TiO₂) decorated by atomic layer deposition. When employing ternary mixtures of ethanol, toluene, and dichloromethane with various compositions, the well-trained 1D-CNN model achieved high accuracy in the predictions, with an average R² score of 0.98 in cross-validation. Unlike prior e-nose studies that use GC-MS only as a benchmark for classification or concentration regression, we reconstruct full 2D GC-MS maps directly from QCM time-series by mapping temporal windows of sensor signals into an invertible PCA latent space of GC-MS and applying the inverse transformthereby linking sensor responses to chemically interpretable peak patterns. These results propose a methodology to bridge the gap between various gas-sensing time-series data and GC-MS molecular data.