Introduction:
Raman spectroscopy enables label-free molecular fingerprinting of serum metabolites, but its clinical use is limited by weak signals from trace analytes and uncontrolled aggregation of gold nanoparticles (Au NPs). Here, we developed a bio–metal hybrid film (BMHF) in which human serum albumin (HSA) acts as a natural capping agent to regulate Au NP assembly and generate reproducible SERS-active nanoclusters for serum-based gastric cancer diagnosis.
Methods:
Serum was mixed with concentrated 80-nm citrate-capped Au NPs at a 1:1 volume ratio and stabilized for 1 h to form a bio–metal hybrid solution. Uniform BMHF substrates were fabricated on SiO substrates by meniscus-dragging deposition (MDD). The coating speed, blade-to-substrate gap, and HSA concentration were optimized using scanning electron microscopy and Raman spectroscopy. FDTD were conducted to assess electromagnetic-field enhancement as a function of the interparticle gap. Raman spectra were collected from serum samples of 50 gastric cancer patients and 50 controls. Spectral features in the 500–1750 cm−1 region were normalized and classified using a deep neural network optimized by Bayesian hyperparameter optimization and five-fold cross-validation.
Results and Discussion: The optimal BMHF was obtained at a coating speed of 5 mm/s, a blade-to-substrate gap of 500 µm, an HSA concentration of 100 nM, and a 1-h stabilization period. HSA suppressed irreversible Au NP aggregation while producing a uniform three-dimensional nanocluster structure with an average Au NP gap of approximately 10 nm. This gap condition was consistent with FDTD showing the greatest effective electric-field enhancement. The BMHF detected serum diluted to 1:100 and rhodamine 6G at M, and it maintained stable Raman responses over 10 days and 200 consecutive measurements. In the independent test set, the neural-network classifier achieved 96.7% sensitivity, 85.6% specificity, and an area under the receiver-operating-characteristic curve of 0.97 for discrimination between gastric cancer and control groups.
Conclusions: The HSA-mediated BMHF provides a stable and reproducible SERS substrate by controlling Au NP assembly and maintaining accessible plasmonic hotspots. Combined with neural-network-assisted spectral classification, this label-free serum platform demonstrated high diagnostic performance for gastric cancer and has potential for blood-based metabolomic screening.