Machine Learning-assisted Raman Spectral Analysis of Serotonin-responsive ssDNA-SWCNT Nanosensor for Improved...

Serotonin (5-hydroxytryptamine, 5-HT) plays critical roles in neuromodulation, yet current detection methods struggle to provide real-time sensing of 5-HT with high sensitivity and selectivity. We previously developed a near-infrared serotonin nanosensor (nIRHT), which consists of ssDNA-wrapped single-walled carbon nanotube that sensitively detects 5-HT. However, nIRHT's fluorescence response cannot discriminate between 5HT and dopamine (DA), limiting its practical applications. In this study, Raman spectroscopy combined with machine learning overcomes this selectivity challenge. G-band spectral features revealed distinct signatures for 5HT versus DA binding to nIRHT, with DA causing greater G-band suppression. We employed differential Raman (ΔRaman) to isolate analyte-specific spectral changes and trained three machine learning models for classification. The random forest model with ΔRaman achieved optimal performance with 95.8% accuracy, significantly outperforming models using raw Raman spectra. This approach showed improved specificity, with negligible responses to acetylcholine, GABA, and glutamate, and achieved a detection limit of 0.1 µM suitable for physiological applications. This Raman-based approach transforms the non-selective nIRHT fluorescence sensor into a platform capable of robust neurotransmitter discrimination, overcoming selectivity issues in single-walled carbon nanotube (SWCNT)-based molecular sensing.