GSH and Halides Directed Controllable Synthesis of Chiral Gold Nanostars for Sensitive Diagnosis of Nephritis Types Using SERS and Transformer Neural Network

Surface-enhanced Raman spectroscopy (SERS) has shown potential for early disease diagnosis via urinary metabolomics, but still faces challenges in achieving stable hot spots and processing complex clinical data. In this study, the preparation of chiral gold nanostars with precisely controllable branch size, number, and sharpness was realized by investigating the effects of l-GSH and CTA<i>X</i> (<i>X</i> indicates halides) on site occupancy, reduction rate, and selective adsorption on crystal facets. Raman spectroscopic characterization using rhodamine 6G (R6G) as a reporter molecule revealed that nanoparticles with fewer branches, larger branch bases, and smoother surfaces exhibited excellent SERS activity, with an analytical enhancement factor (AEF) of 5.49 × 10⁶ in the nonresonance region, which was also supported by finite-difference time-domain (FDTD) calculations. The hierarchical self-assembly of gold nanostructures was achieved by introducing a hydrophobic template to control the substrate size, significantly improving the reproducibility of SERS substrates. Combined with a Transformer neural network, this study achieved 99.94% diagnostic accuracy in distinguishing urine samples from healthy individuals and patients with acute interstitial nephritis and nonacute interstitial nephritis, demonstrating the potential of SERS and deep learning in clinical diagnostics.