Early identification and survey of biological analytes, including neurotransmitters and hormones, are essential for early disease diagnosis, tailored treatment, and immediate physiological evaluation. Dysregulations in dopamine, cortisol, oxytocin, and serotonin are associated with a wide array of illnesses, encompassing neurodegenerative, psychiatric, cardiovascular, and metabolic conditions. In this study, we present an AI-assisted sensing platform utilizing reduced graphene quantum dots (RGQDs) for the multiplex detection of analytes (dopamine, serotonin, cortisol, and oxytocin) by spectral fingerprint identification generated due to analyte-binding-facilitated fluorescence quenching. RGQDs exhibit photostable near-infrared emission, allowing for signal detection through the layers of biological tissue and remarkable in vivo biocompatibility, altogether enabling their utility in implanted or wearable diagnostic systems. In contrast to conventional sensors that depend on molecular specificity, this study employs cross-reactive RGQDs, whose fluorescence is influenced by noncovalent interactions with aforementioned analytes, evaluated via the conformer–rotamer sampling tool (CREST). This theoretical model of interactions of analyte with the RGQD surface computes optimized dimer geometries, binding energies, and π–π stacking configurations, revealing analyte-specific noncovalent binding patterns that enable selective sensing. Artificial intelligence models are developed and trained to recognize spectral variations unique to each analyte. Through unsupervised dimensionality reduction and nonlinear classification (UMAP combined with a random forest model), the system distinguishes analytes using three key spectral features: continuum-removed emission area, first-derivative amplitudes, and wavelet detail energy at level 4, achieving an overall discrimination accuracy of 91.2% and a macro F1 score of 91.5%. This approach facilitates multiplexed nonspecific sensing of four critical hormones/neurotransmitters without the need for designing complex analyte-specific surface groups. This work beneficially merges material innovation with machine learning to provide a cost-efficient, noninvasive, and scalable diagnostic platform, aimed to enhance healthcare accessibility in both clinical and underserved environments.
Sharma et al. (Mon,) studied this question.