To address these challenges, we have developed a photoacoustic (PA) sensing platform based on a whispering-gallery-mode (WGM) optical microresonator integrated with microfluidic channels. WGM resonators support ultra-high quality (Q) optical modes and dramatically enhance light-matter interactions, enabling highly sensitive detection of nano- and micro-scale objects, such as cells, in microliter-scale samples. Optical WGM resonators have demonstrated Q-factors at least two orders of magnitude higher than other kinds of optical resonators. In a WGM optical micro-resonator, light propagates along the curved surface of a circular boundary over 106 times through total internal reflections, which significantly enhances light-matter interactions for sensing applications. Recently WGM sensors have emerged as a frontrunner in biochemical analysis for the detection of small objects such as nanoparticles, viruses, and biomolecules. PA spectroscopy, which uses a pulsed laser to excite objects, such as cells, and generate acoustic waves through transient thermoelastic expansion, provides rich spectral signatures that reflect the optical, thermal, and mechanical properties of target cells. By integrating PA spectroscopy with WGM sensors, we demonstrate a powerful technique for label-free, and non-invasive characterization of microbial populations. AI plays a critical role in analyzing mixed populations and signatures between species in the PA signals. We employ machine learning models, with feature-extraction pipelines and supervised classification algorithms, to decode the spectral information embedded in the PA signals. Unlike conventional signal processing approaches, this AI-enhanced approach can identify subtle, multidimensional patterns that are difficult to detect manually or with traditional metrics. This makes it possible to distinguish between cell types, quantify relative abundances, and monitor dynamic changes over time in microbial communities.
In our demonstration involving mixtures of wild-type and engineered yeast, we collected thousands of PA spectra per sample to ensure statistical rigor. An AI model trained on these data achieved 91.8% accuracy in classifying strain composition and predicting mixture ratios across a wide range of concentrations. Our technique demonstrated the ability to detect early-stage contamination and monitor co-culture of yeast and cyanobacteria—tasks not achievable with conventional OD-based methods.
This integrated sensing approach that combines optics, microfluidics, photoacoustics, and machine learning offers a powerful tool for synthetic biology. It offers superior sensitivity, robust performance across a wide range of sample concentrations, and an improved detection limit compared to conventional methods. These advancements highlight the sensor's potential for real-time, label-free, and non-invasive monitoring of microbial populations, making it a valuable tool for bioprocessing optimization and contamination detection in complex biological environments, where continuous and precise control is essential.