Data-driven nanomechanical sensing: specific information extraction from a complex system

Abstract

Abstract Smells are known to be composed of thousands of chemicals with various concentrations, and thus, the extraction of specific information from such a complex system is still challenging. Herein, we report for the first time that the nanomechanical sensing combined with machine learning realizes the specific information extraction, e.g. alcohol content quantification as a proof-of-concept, from the smells of liquors. A newly developed nanomechanical sensor platform, a Membrane-type Surface stress Sensor (MSS), was utilized. Each MSS channel was coated with functional nanoparticles, covering diverse analytes. The smells of 35 liquid samples including water, teas, liquors, and water/EtOH mixtures were measured using the functionalized MSS array. We selected characteristic features from the measured responses and kernel ridge regression was used to predict the alcohol content of the samples, resulting in successful alcohol content quantification. Moreover, the present approach provided a guideline to improve the quantification accuracy; hydrophobic coating materials worked more effectively than hydrophilic ones. On the basis of the guideline, we experimentally demonstrated that additional materials, such as hydrophobic polymers, led to much better prediction accuracy. The applicability of this data-driven nanomechanical sensing is not limited to the alcohol content quantification but to various fields including food, security, environment, and medicine.