Rapid and Differential Diagnosis of Sepsis Stages Using an Advanced 3D Plasmonic Bimetallic Alloy Nanoarchitecture-Based SERS Biosensor Combined with Machine Learning for Multiple Analyte Identification

Abstract

Rapid and accurate differential diagnosis of infections, sepsis, and septic shock is essential for preventing unnecessary antibiotic overuse and improving the chance of patient survival. To address this, a 3D gold nanogranule decorated gold-silver alloy nanopillar (AuNG@Au-AgNP) based surface-enhanced Raman scattering (SERS) biosensor is developed, capable of quantitatively profiling immune-related soluble proteins (interleukin three receptor, alpha chain: CD123, programmed cell death ligand 1: PD-L1, human leukocyte antigen-DR isotype: HLA-DR, and chitotriosidase: ChiT) in serum samples. The 3D bimetallic nanoarchitecture, fabricated using anodized aluminum oxide (AAO), features a uniform structure with densely packed nanogaps on the heads of Au-Ag alloy nanopillars, enabling fast, simple, and replicable production. The proposed biosensor achieves accurate results even with low detection limits (4-6 fM) and high signal consistency (relative standard deviation (RSD) = 1.79%) within a one-step multi-analytes identification chip with a directly loadable chamber. To enhance the diagnostic performance, a support vector machine (SVM) based machine learning algorithm is utilized, achieving 95.0% accuracy and 95.8% precision in classifying healthy controls, infections with and without sepsis, and septic shock. This advanced 3D plasmonic bimetallic alloy nanoarchitecture-based SERS biosensor demonstrates clinical usefulness for sepsis diagnosis and severity assessment, providing timely and personalized treatment.

Keywords: 3D bimetallic alloy nanoarchitecture; all‐in‐one multiplex detection; differential diagnosis of sepsis; soluble protein; surface‐enhanced Raman scattering (SERS).

Figure 1

Fabrication of the 3D AuNG@Au‐AgNP Substrate. A) Scheme of the fabrication process. B, C) SEM images showing the top view of (B) the Au‐Ag NPs and (C) the fabricated 3D AuNG@Au‐AgNP SERS substrates after metal sputtering. D, E) Numerical simulations depicting the electromagnetic near‐field distribution for (D) Au‐Ag NPs without Au nanogranules and (E) the 3D AuNG@Au‐AgNP SERS substrates. F) Simulated maximum SERS enhancement factor, proportional to the fourth power of the electromagnetic field magnitude (|E|⁴). G) SERS spectra of IgG‐cyanine 3 and SERS signal intensity at 1321 cm⁻¹ (characteristic of Cy3) for 3D AuNG@Au‐AgNP substrates with varying Au nanogranule sizes, using 10 nM IgG‐cyanine 3 as a model analyte.

Figure 2

Schematic of soluble protein detection in blood using the 3D AuNG@Au‐AgNP based SERS biosensor with the 3D‐printed multiplex chip and direct‐loadable chamber. This biosensor allows direct sample incubation using an antigen‐antibody sandwich assay to detect target soluble proteins. Blood and serum samples were centrifuged with a benchtop centrifuge for 10 min. The multiplex chip containing four 3D AuNG@Au‐AgNP substrates attached with double‐sided tape was incubated during centrifugation. Following incubation, SERS signals were measured using the 3D‐printed direct‐loadable chamber.

Figure 3

Evaluation of sensing performance of 3D AuNG@Au‐AgNP‐based SERS biosensor A) Averaged SERS spectra (dashed line) with standard deviations (shaded area) for PD‐L1 (Cy3) at 100 nM, measured across a 50 × 50 µm area (2500 points) on a single SERS biosensor. B) Selectivity assessment of the 3D bimetallic alloy nanoarchitecture‐based SERS biosensor through SERS intensity measurements at specific Raman reporter peaks, validated for target soluble proteins (CD123, PD‐L1, ChiT, HLA‐DR) using a confusion matrix. C) Sensitivity analysis of the 3D AuNG@Au‐AgNP‐based SERS biosensor: SERS spectra over a concentration gradient and corresponding plot of specific peak intensities (1131 cm⁻¹, 1240 cm⁻¹, 1321 cm⁻¹, and 1516 cm⁻¹ for CD123, HLA‐DR, PD‐L1, and ChiT, respectively) against soluble protein concentrations in spiked human serum. Linear regression analysis was performed to assess the correlation between soluble protein concentration and specific peak intensity. Error bars represent standard deviations from 30 measurements.

Figure 4

Clinical significance evaluation using the 3D bimetallic alloy nanoarchitecture‐based SERS biosensor and dual track detection approaches with machine learning. Schematic illustration of dual track detection approaches for differential diagnosis among clinical groups, utilizing expression level ratios and SVM algorithm analysis, with all‐in‐one incubation for complex SERS spectral measurement enabling multiple analyte identification.

Figure 5

Clinical significance evaluation for differential diagnosis among clinical groups utilizing expression level ratios using the 3D bimetallic alloy nanoarchitecture‐based SERS biosensor. Box plots showing the expression level ratios of A) PD‐L1/CD123 and B) ChiT/HLA‐DR, comparing healthy controls and symptomatic groups (n = 10 per group) for sepsis diagnosis and severity prediction, samples undetected by ELISA represented in skyblue rectangles (***p < 0.001, **p < 0.01, *p < 0.1 ns; not significant).

Figure 6

A) Confusion matrix displaying the diagnostic accuracy of SVM classifiers in distinguishing healthy control group (n = 10) from symptomatic groups (n = 30). B) Differential diagnosis of clinical unknown patients (n = 3) using SVM algorithm‐based machine learning.