Shape-Encoded Hydrogel Sensor Particles Enable Multiplex Odorant Detection Through Deep-learning Classification

Simultaneous detection of multiple odorants is a major challenge in the development of portable, cell-based biohybrid sensors, primarily due to the difficulty of distinguishing between different sensor cell types. Here, a strategy that encodes odorant sensor cell types using the shape of hydrogel particles, enabling shape-based identification through deep learning is reported. Each particle shape corresponds to a unique sensor cell type expressing a distinct odorant receptor (OR). A convolutional neural network is trained to classify these shapes with high accuracy, and the resulting shape identification scheme is applied to time-lapse fluorescence images of mixed particles exposed to single odorants. This enabled reliable assignment of particle identity and extraction of shape-specific fluorescence signals. Distinct odorant-dependent responses are observed, consistent with the known ligand specificities of the corresponding ORs. While this study focuses on individual odorants, the shape-based approach provides a position-independent, scalable method for multiplexed odorant detection. This framework supports the development of compact, high-throughput biohybrid sensors for safety, environmental monitoring, and diagnostic applications.