Infrared nanoplasmonic metasurfaces augmented by artificial intelligence for universal biosensing

Nanoplasmonic metasurfaces have shown outstanding light-matter interaction enhancement capabilities, leading to their emergence as powerful platforms for highly sensitive biospectroscopy. Metasurface-enhanced biospectroscopy offers unprecedented opportunities for biological studies, and its full potential remains to be unleashed. Mid-IR metasurfaces, in particular, are very promising because they can act as amplifiers of fingerprint-like molecule vibrations, which are plentiful in this rich spectral range. In this thesis, we develop novel nanoplasmonic designs coupled with custom microfluidics and artificial intelligence-based data analysis models to demonstrate real-time, label-free, chemically-specific, and non-destructive monitoring of biomolecules and their interactions in aqueous media. Our first nanoplasmonic design combines optimized grating order-coupled nanoantenna arrays with protein-accessible nanogaps to enable the high sensitivity monitoring of proteins and their three-dimensional structures in aqueous media. The engineered nanoantennas reach electric field intensity enhancements of up to five orders of magnitude and provide chemically specific detection of proteins and their secondary structures down to picograms and nanograms per milliliter, respectively. In the next part of the thesis, we develop multiresonant metasurfaces to monitor interactions between biomolecules with vibrational fingerprints in different parts of the mid-IR range. Our first effort focuses on developing a nanoplasmonic design for simultaneous monitoring of both proteins and lipid molecules. Lipids are another important class of biomolecules as they are the building blocks of biological membranes, and lipid-protein interactions are at the core of many cellular processes. New analytical tools for their study in water and at the monolayer level are of fundamental importance. Therefore, we introduce a dual-resonant nanoplasmonic design coupled to machine learning-based data analysis to overcome current sensor challenges. We apply our technology to a dynamic system involving synaptic vesicle mimics and demonstrate that we can resolve complex mass-preserving biological interactions in real-time. This is a remarkable feat that traditional non-chemically specific analytical measurement tools such as surface plasmon resonance or quartz crystal microbalance spectroscopy could not achieve. In the final part of the thesis, we develop yet another multiresonant design for broadband coverage of the whole mid-IR range. We couple our sensor to a deep learning model to resolve a dynamic biological system including all major classes of biomolecules simultaneously. Specifically, we resolve the toxic peptide-induced release of carbohydrates and nucleotides from exosome-like bionanoparticles.