Gold- or Silver-Nanoparticle SERS Platforms for Plasma-Based Diagnostics and AI-Driven Analysis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This review article provides a comprehensive overview of gold- or silver-nanoparticle SERS platforms for plasma-based diagnostics, integrating AI-driven analysis. The manuscript covers the fundamental principles of SERS, various SERS-active nanoparticles, biological matrices, disease diagnostics applications, and data analysis strategies. The authors also discuss future perspectives, challenges, and opportunities in this rapidly evolving field. The topic is highly relevant and timely, addressing a critical area in biomedical diagnostics and the application of advanced analytical techniques. Despite its breadth, the manuscript currently reads more as an extensive literature compilation than a critical review. Several structural, formatting, and scientific issues need to be addressed. There are also significant issues related to organization, consistency of scope, formatting, figure referencing, critical analysis, and language quality that should be addressed before the manuscript is suitable for publication. I therefore recommend major revision before the manuscript can be considered for publication. I recommend that the authors address the following comments fully and integrate their responses directly into the manuscript to warrant publication.

1) There are multiple unresolved formatting problems throughout the manuscript: Figure numbering is inconsistent, with multiple figures labelled “Figure 1, some captions are excessively long, and several figures appear to be reproduced from prior publications; permissions and formatting should be carefully verified. These issues must be corrected before publication consideration.

2) The manuscript is very long and at times difficult to follow. Some sections should be merged or streamlined

3) The authors state that this review specifically focuses on plasma-based diagnostics using AuNP/AgNP SERSA platforms. However, the manuscript currently provides limited critical differentiation from prior SERS review. The authors should explicitly clarify: What unique perspective this review contributes beyond existing SERS diagnostic reviews, and how plasma-based diagnostics differ analytically from serum or other biological matrices

4) The title and introduction indicate that the review focuses on plasma-based diagnostics. However, large sections discuss urine, saliva, tears, sputum, feces, tissues, cerebrospinal fluid, bacteria, and other biological matrices. While these examples are informative, they dilute the stated focus of the review. The authors should clearly justify the inclusion of these sections or substantially condense them. Greater emphasis should be placed on plasma-specific challenges, including protein corona formation, high-abundance plasma proteins, matrix interference, sample preprocessing requirements, and clinical validation studies using plasma.

5) The manuscript predominantly summarizes published studies but provides limited critical evaluation. The review would be significantly strengthened by comparing AuNP versus AgNP performance, comparing label-free versus tagged SERS approaches, discussing reproducibility issues quantitatively, highlighting limitations of reported studies, comparing diagnostic performance across disease categories, and discussing reasons for discrepancies between studies. Could the authors add additional references related to different pathways of seed-mediated growth of single-shell and multi-shell gold and silver nanoparticles.

6) The title explicitly includes “AI-Driven Analysis,” yet the AI discussion appears relatively disconnected from the diagnostic examples. I recommend the authors to include a description of different machine learning algorithms reported in literature for Raman spectroscopy data analysis (eg – PSE-LR - peak sensitive elastic net regularization, RamanSpy, PyFasma, SSNet, RADAR)

7) Several sections describe numerous studies individually, making it difficult for readers to compare results. The manuscript would benefit from summary tables containing disease, biomarker, plasma/serum sample type, SERS substrate, detection strategy, machine-learning method, sensitivity/specificity/AUC, and limit of detection.

8) The manuscript repeatedly states that SERS has strong diagnostic potential but has not yet achieved widespread clinical adoption. The authors should provide a better discussion on this 

9) One of the issues with SERS is mitigating inherent noise when using plasmonic nanoparticles alone. Could the authors elaborate on different pathways to manage the noise in SERS with references (eg – spread spectrum SERS allows label-free detection of attomolar neurotransmitters, Noise management of surface-enhanced Raman spectroscopy using two-dimensional materials, Enhancement of the signal-to-noise ratio in fiber-optics based SERS detection by rough-cutting the end surface)

 

Author Response

We thank the reviewer for their revision of our manuscript and suggestions given. Following, our point-by-point answers in blue text, with corrections highlighted in blue in the revised manuscript.

• There are multiple unresolved formatting problems throughout the manuscript: Figure numbering is inconsistent, with multiple figures labelled “Figure 1, some captions are excessively long, and several figures appear to be reproduced from prior publications; permissions and formatting should be carefully verified. These issues must be corrected before publication consideration.

We thank the reviewer for alerting us for these matters, that we have resolved in the revised manuscript, namely:

1. We have carefully reviewed figure numbering, and believe it is consistent and correct in the revised version of the ms.
2. We have shortened, keeping only essential information, the captions in figures 5 and 8 (in the new numbering), as well as in figure 10.

• For figures reproduced from prior publications; all necessary permissions and formatting were carefully verified.
• The authors state that this review specifically focuses on plasma-based diagnostics using AuNP/AgNP SERSA platforms. However, the manuscript currently provides limited critical differentiation from prior SERS review. The authors should explicitly clarify: What unique perspective this review contributes beyond existing SERS diagnostic reviews, and how plasma-based diagnostics differ analytically from serum or other biological matrices.

We thank the reviewer for their valuable suggestion, we have expanded the introductory paragraphs in Section 1 to explicitly articulate the unique scope of this review, namely, the convergence of blood plasma as the sole biological matrix, AuNP/AgNP as the plasmonic platforms, and AI-driven spectral analysis. Distinguishing it from prior SERS reviews that address broader matrices or omit systematic chemometric discussion. We have also added a paragraph clarifying how plasma presents distinct analytical challenges relative to serum (absence of clotting factors, preserved coagulation proteins) and other biofluids, reinforcing the clinical rationale for this focused scope. It is also implicitly addressed in Sections 5 and Section 7.

In the Introduction by scope. The diverse reviews on SERS are expansive with regards to the biological matrix. This review limits the matrix to plasma only.

Despite the volumes of the available literature detailing SERS diagnostics, includ-ing  several comprehensive reviews of SERS methodology [6, 7, 44, 56], no existing re-view focuses specifically on plasma as the biological matrix, gold (Au) and silver (Ag) nanoparticles (NPs) as the plasmonic substrates, and chemometric methods encom-passing the diverse traditional statistics and associated artificial intelligence (AI) algo-rithms for SERS spectra interpretation. A combination that represents different strate-gies researchers are exploring to address SERS assay challenges arising from the sub-strates, the biological matrix, and the data interpretation platforms. The scope of this review is therefore defined by three parameters, namely, the biological matrix (blood plasma), nanoparticle substrate (AuNP and AgNP, including hybrid architectures), and SERS data interpretation framework (including linear chemometrics, machine learning (ML) and explainable AI). Figure 1 presents a schematic overview of the scope and main findings of the literature reviewed.

 

(Figure 1)

 

The sections of this review are organised to provide a brief introduction to the principles and mechanisms of SERS enhancement and the direct/indirect detection strategies in Section 2. An introduction of AuNP and AgNP as SERS-active materials and their properties is highlighted in Section 3. Section 4 briefly highlight biomarkers in biological matrices, with a focus on blood plasma and matrix-derived challenges such as the protein corona formation, coffee ring effect, and SERS result variability due to random aggregation when using colloidal suspensions. Section 5 provides an over-view of clinically relevant SERS research using plasma as the biological matrix. Or-ganised by detection modality that progresses from label-free single-mode strategies through to labelled multiplexed immunoassays and multimodal biosensing systems across a range of oncologic and non-oncologic target analytes. A detailed presentation of SERS data analysis pipeline is covered in Section 6. From spectral preprocessing through dimensionality reduction to machine learning classification and explainable AI methods. Section 7 briefly discuss the challenges limiting the translation of SERS from the research laboratory to clinical utility. It also highlights opportunities could determine the trajectory of the SERS research field. Conclusion are presented in Section 8.

• The title and introduction indicate that the review focuses on plasma-based diagnostics. However, large sections discuss urine, saliva, tears, sputum, feces, tissues, cerebrospinal fluid, bacteria, and other biological matrices. While these examples are informative, they dilute the stated focus of the review. The authors should clearly justify the inclusion of these sections or substantially condense them. Greater emphasis should be placed on plasma-specific challenges, including protein corona formation, high-abundance plasma proteins, matrix interference, sample preprocessing requirements, and clinical validation studies using plasma.

We thank the reviewer for their valuable suggestion. Accordingly, non-plasma biofluid sections (urine, saliva, tears, etc.) have been removed and the section dedicated only to blood plasma. Also, a section 4.2.3 has been added that briefly highlights the protein corona and matrix interference from plasma constituents in SERS assays. We acknowledge that sections discussing urine, saliva, tears, sputum, and other matrices extend beyond the stated plasma focus, and we have substantially condensed these sections, retaining only brief comparative references where they directly contextualise plasma-specific findings. Greater emphasis on plasma-specific challenges including protein corona formation, high-abundance plasma protein interference, sample preprocessing, and clinical validation has been added to Section 4.1 (also in Table 4.1), Sections 4.2.3. the challenges are already highlighted throughout Section 5, and specifically in Section 7.1.2.

 

4.1 Biological Matrices

Biological matrices are the source of biomarkers targeted in diagnostic SERS analyses, ranging from complex, protein-rich biofluids such as blood plasma and cerebrospinal fluid (CSF), through cellular and tissue specimens, to structurally heterogeneous solid matrices including bone, tooth enamel, and keratinous appendages such as hair and fingernails [5], [103]. As repositories of vital anatomical, physiologic, and pathologic information, their applicability extends beyond disease diagnostics to include forensic and toxicological investigations for medical or legal purposes [5], [104]. Solid specimens like bones, teeth, fingernails, and hair, widely known to endure environmental degradation, are mostly exploited in forensic SERS assays for crime investigations [95], [97]. Health status from normal/abnormal levels of these sweat-based biomarkers can be inferred via regular SERS assays [104], [105], or novel wearable SERS sensors [106], [107]. Human tears has served as clinical samples for SERS based detection of Alzheimer’s [109], and diabetes [110] among other diseases. Sputum and saliva have been samples in diagnostic SERS for pathogenic infections [111]; inflammatory agents [87]; periodontal diseases [112]; systemic autoimmune diseases [19], [113]. SERS has probed urine samples for evaluating kidney function through the variation in detected biomarker content levels [114], [115], [116].  The Cerebrospinal fluid (CSF) has been used to detect pathogenic infection of the CNS [117], [118], neoplasms [119] and other disorders of the CNS using a variety of AuNP/AgNP substrates. Given the complexity of biological matrices, together with the variation in biomarker quality and quantity, an understanding of biological matrix properties and composition informs their suitability as samples for different analytical procedures [105] a summary of different biological matrices, their complexity, advantage and disadvantages is highlighted in Table 4.

Of all the biological matrices, blood-based samples (whole blood, plasma, serum) are the most used across all disease diagnostics [102], [103], [120]. Whole blood comprising of blood cells (erythrocytes, leucocytes, and thrombocytes) suspended in plasma and contains biomarkers from almost all the body organs [121](39,90). Serum is whole blood completely separated from blood cells. Clinical plasma is the liquid blood portion devoid of red and white blood cells, while retaining the coagulating factors, hence has higher protein content than serum [103], [122]. While serum is less rich in proteins, hence have a reduced matrix effect on metabolite adsorption, plasma contains the full complement of blood protein. There is no consensus on which is better as a clinical biological matrix [37], [122]. Taken together, they contain the most comprehensive biomarker components relevant to an endless list of diseases for diagnostics, biomarker discovery and therapeutic drug monitoring [121], [123]. SERS investigation using whole blood samples [124], [125], [126], serum [127], [128], [129], and plasma [84], [130], [131] dominates past and contemporary research landscape.

Human blood plasma and serum are the most studied bodily fluids for disease diagnosis, biomarker discovery and therapeutic drug monitoring [106], [107], [108]. As a connective tissue circulating all over the human body, plasma composition is continuously equilibrated with the extracellular fluid of virtually every tissue compartment, providing information about cell turnover, inflammation, and antioxidant capacity [112]. In short, when these processes are perturbed in virtually any disease, plasma reflects the complete physiologic and pathologic state of the human body system, making blood test using plasma a routine in clinical diagnosis [112], [113]. While blood plasma composition is predominantly water (≥ 90%), with over 114,000 known metabolites at varying concentration level (< 1 nmol/L to mmol/L), minerals, organic substances and gas [109], [110], [111]. Its molecular components is dominated by proteins of different molecular weight fractions (albumin, globulins, fibrinogen and a thousand others) [109], [110], together with constituents like carbohydrates, lipids and amino acids. It is obvious that all the biomarkers earlier mentioned in the preceding section are present in plasma. And their fingerprint can be derived from SERS measurements, given their Raman spectrum are in the range of 400—2000 cm ^-1 wavenumbers, where bond vibrations at 470 – 1200 cm^-1 are associated with carbohydrates; 980, 1080 and 1240 cm^-1 are associated with some metabolites and nucleic acid phosphate groups; 1500 – 1700 cm^-1 are associated with proteins; and higher wave numbers (2700 3500 cm^-1) are attributed to CH, NH, and OH stretching in protein and lipids [45]

 4.3. Challenges of SERS Measurements in Blood Plasma

The principal limitation of the label-free approach in plasma analysis is the susceptibility of spectral quality to interference from the matrix itself. These challenges can be classed into four, namely, the protein corona; spectral interference; Nanoparticle instability and degradation; and pre-analytical variability.

The introduction of NPs into a plasma-containing medium results in rapid adsorption of plasma protein onto the NP surface, forming the protein corona that alters the NP physicochemical properties alongside influencing the NP behaviour in plasma. Based on kinetics, two corona layers, the hard corona with higher adsorption affinity forms an inner layer of slow dissociating proteins, while the soft corona comprised of proteins with lower binding affinity form the outer layer that is in constant flux [109], [111]. The formation of the protein corona has three analytically catastrophic consequences for SERS. First, it physically displaces analyte molecules from the plasmonic hot-spot region at the nanoparticle surface, reducing their proximity to the maximum near-field enhancement zone and thus suppressing their SERS signal. Secondly, the corona proteins themselves, particularly albumin, fibrinogen, and other high abundance protein fractions also generate SERS signals when they adsorb directly onto the NP surface. And the intensity of these background signals can sometimes mask the signals of low-abundance target biomarkers. Thirdly, the corona alters the nanoparticle surface charge, reducing electrostatic repulsion and promoting nanoparticle aggregation with consequent loss of SERS reproducibility. Given that the protein corona composition is not static, constantly evolving based on the Vroman effect in which initially adsorbed high-abundance proteins are progressively displaced by lower-abundance proteins of higher affinity as the equilibration proceeds. This Vroman effect introduces spectral drift during SERS measurements performed at different timeframes, leading to significant measurement irreproducibility in plasma SERS protocols.

The second challenge, spectral interference in plasma SERS arises when molecular components of the matrix generate Raman bands that overlap with, mask, or are misidentified as the target analyte signal. Unlike fluorescence background which can be partially removed by baseline correction algorithms, spectral interference from structurally defined plasma molecules produces sharp, reproducible Raman bands at specific wavenumber positions that are indistinguishable from genuine analyte signals unless their molecular origin is known. Albumin is the dominant spectral interferent in plasma SERS, generating intense bands that coincide with diagnostically important regions of the SERS spectrum for other target analytes [101], [109]. Also, haemoglobin released from lysed erythrocytes in haemolytic plasma specimens generates several intense Raman bands arising from its porphyrin ring and iron–histidine coordination environment. These bands can dominate the SERS spectrum and saturate the CCD detector, rendering the measurement analytically valueless. Another interferant, lipoproteins contribute a characteristic band which overlap with carbohydrate and nitrogenous bases in the same region.

The third challenge is colloidal stability of SERS nanoparticles in plasma when NPs are critically compromised by the high ionic strength of the plasma environment. Citrate-stabilised AuNPs suspensions, which rely on electrostatic repulsion between negatively charged surface ligands to prevent aggregation, are stable in deionised water, but undergo rapid aggregation in physiological saline where the Debye screening length become insufficient to maintain an electrostatic repulsion energy barrier against the van der Waals attractive force between particles. This critical aggregation produce highly variable hot-spot distributions whose SERS signal intensities fluctuate between measurements, eliminating any possibility of quantitative calibration. For AgNPs, an additional metal-specific degradation mechanisms occurs in plasma. Oxygen and plasma constituents that contain sulphur react with the Ag surface to form SERS-inactive compounds. The SERS activity of non-passivated AgNPs become rapidly extinguished in plasma, thereby confounding SERS measurements[108], [111].

Lastly, pre-analytical variability where variation in plasma composition arises from sample collection, processing, and storage conditions rather than from genuine biological differences between subjects is a most often overlooked source of irreproducibility in plasma SERS measurements. The sensitivity of SERS is such that alterations in the molecular composition of the plasma matrix seriously affect the SERS spectral profile. Also, the choice of anticoagulant affects plasma SERS spectra. EDTA (ethylenediaminetetraacetic acid), heparin, and citrate anticoagulants are the three most commonly used agents in clinical plasma collection, and each introduces unique compositional changes. EDTA chelates divalent cations (Ca2+, Mg2+), altering the protein conformation of calcium-dependent plasma proteins and changing their nanoparticle adsorption behaviour. The EDTA molecule also generate SERS signals when they adsorb onto Au and Ag NPs surfaces. Heparin binds to several plasma proteins and alters the protein corona composition relative to EDTA or citrate plasma. While plasma can be diluted by the citrate anticoagulant present in blood collection tubes, hence reduces all analyte concentrations and alters electrolyte balance. Repeated freeze–thaw cycle induces protein denaturation and aggregation in plasma, alters lipoprotein structure, and releases intracellular contents from intact residual cells or platelet fragments in the plasma sample [135]. The cumulative structural alteration in plasma due to repeated freeze–thaw cycles can cause significant spectral sift during SERS measurements. Processing delay between time of blood collection to plasma processing allows ongoing metabolic activity of residual leukocytes and platelets in whole blood, which alters glucose, lactate, and cytokine concentrations in a time-dependent manner. These pre-analytical variables are a primary cause of false positive and false negative classifications.

The four challenges identified above often reinforce each other, for example, protein corona formation promotes NP aggregation and introduces spectral interference through adsorbed protein SERS bands simultaneously. Similarly, NP aggregation creates variable hotspot distributions that amplify the impact of any residual spectral interference, while pre-analytical variability alters the protein corona composition, the degree of nanoparticle aggregation, and the intensity of spectral interferences in a sample-specific manner.

Effective plasma sample preparation for SERS therefore requires a multi-step, integrated approach that addresses all four challenge domains in a defined sequence: first, pre-analytical standardisation to control the composition of the starting plasma; second, matrix simplification to reduce the concentration of dominant interferent proteins and stabilise the nanoparticle system; third, nanoparticle surface engineering to resist corona formation and degradation; and fourth, measurement standardisation to compensate for residual inter-sample variation[136], [137]. In response, these matrix-derived challenges have motivated the development of sample pretreatment strategies  including molecular weight filtration, pH adjustment, and dilution, intended to selectively reduce HMWF interference without substantial loss of diagnostically relevant spectral information [46], [81], [138]. Their comparative performance is examined in Section 5 in the context of specific diagnostic applications.

 

• The manuscript predominantly summarizes published studies but provides limited critical evaluation. The review would be significantly strengthened by comparing AuNP versus AgNP performance, comparing label-free versus tagged SERS approaches, discussing reproducibility issues quantitatively, highlighting limitations of reported studies, comparing diagnostic performance across disease categories, and discussing reasons for discrepancies between studies. Could the authors add additional references related to different pathways of seed-mediated growth of single-shell and multi-shell gold and silver nanoparticles.

We thank the reviewer for pointing this out and have strengthened the critical analysis throughout Section 5 by explicitly comparing AuNP and AgNP diagnostic performance, contrasting label-free and labelled SERS approaches in terms of reproducibility and LOD, and discussing reasons for inter-study discrepancies, including substrate heterogeneity, sample size limitations, and varying preprocessing protocols.

The entire Section 5 has been reorganised into specific subsections aimed to enhance coherence and readability as shown from the Table of Contents:

 

5. SERS, Plasma and AuNP/AgNP Nanostructures Utility in Disease Diagnostics. 25

5.1. Single-Mode Single/Multiplex Analyte Detection. 26

5.1.1. Colloidal Substrate Strategies in Oncological Diagnostics. 26

5.1.2. Solid-Support and Anisotropic Substrate Strategies. 30

5.1.3. Metabolic and Non-Oncological Targets. 31

5.2. Single-Mode Detection: Labelled and Indirect Approaches. 32

5.2.1. Single-Analyte Sandwich Immunoassays. 32

5.2.2. Multiplex Labelled Detection. 34

5.2.3. Combined Label-Free Profiling with Internal Standards: A Methodological Hybrid. 36

5.3. Multi-Mode Detection Strategies. 37

5.3.1. SERS Integrated with Thermoplasmonic Detection: Cardiac Troponin I 37

5.3.2. SERS Integrated with Electrochemistry and Acoustofluidics: Alzheimer's Disease Biomarker Detection. 38

5.3.3. SERS Integrated with Colorimetry and Lateral Flow: Traumatic Brain Injury Detection. 38

 

• Could the authors add additional references related to different pathways of seed-mediated growth of single-shell and multi-shell gold and silver nanoparticles.

We thank the reviewer for their valuable suggestion. We have addressed it in Section 3.5.:

In the seed-mediated synthesis route, in contrast to fabricating NPs from kinetically controlled growth, a more controlled tuning of NP size and shape is achieved by using already synthesised NPs (seeding solution) of suitable size are used as templates in a reaction mixture (growth solution) that also contains reducing agents and capping ligands. Since it is energetically favourable for NPs to form on the seeds, a highly monodispersed final metal NPs with a narrow size range is achieved [84], [85]. The high tunability of the seed-mediated growth synthesis route has made the process suitable for the fabricating single-shell NPS, Au@Ag NPs [86], [87]; and multi-shell NPs [88], [89].

• The title explicitly includes “AI-Driven Analysis,” yet the AI discussion appears relatively disconnected from the diagnostic examples. I recommend the authors to include a description of different machine learning algorithms reported in literature for Raman spectroscopy data analysis (eg – PSE-LR - peak sensitive elastic net regularization, RamanSpy, PyFasma, SSNet, RADAR) Add to Section 6.1.1 Typical Pre-processing Pipelines

We thank the reviewer for this valuable suggestion. We agree that the original manuscript did not sufficiently connect the discussion of AI methodologies with the diagnostic applications presented throughout the review. To address this issue, we expanded the computational sections to include examples of spectroscopy-specific machine learning and deep learning frameworks that have been developed for Raman and SERS data analysis.Specifically, we added discussion of Peak Sensitive Elastic Net Regularization (PSE-LR) as a spectroscopy-oriented regularization strategy  that preserves diagnostically relevant Raman peaks while reducing model dependence on spectral background noise. We also expanded the deep learning section to include Spectra Segmentation Networks (SSNet) and RADAR, highlighting their role as end-to-end architectures designed for one-dimensional spectral data and automatic feature extraction without manual peak selection. In addition, the preprocessing section now discusses open-source platforms such as RamanSpy and PyFasma, which provide standardized and reproducible workflows for spectral preprocessing, normalization, and machine learning integration.

 

These additions were incorporated into Sections 6.1.1 (Typical Pre-processing Pipelines), 6.3 (Machine Learning Models for Spectroscopic Classification), and 6.3.3 (Deep Neural Networks), respectively.

 

After the discussion of baseline correction, smoothing, normalization and preprocessing workflows:

The increasing scale of clinical SERS studies has also highlighted the need for standardized and reproducible preprocessing workflows. To address this challenge, open-source spectroscopy toolboxes such as RamanSpy [*] and PyFasma [**] have emerged as computational backbones for spectral analysis, providing unified implementations of baseline corrrection, normalization, denoising, visualization and machine learning integration. By reducing variability introduced by custom preprocessing pipelines, these frameworks facilitate reproducible analysis across research groups and clinical sites, supporting the development of large-scale multicentre diagnostic studies.

[*] Georgiev, D., Pedersen, S. V., Xie, R., Fernández-Galiana, Á., Stevens, M. M., & Barahona, M. (2024). RamanSPy: An open-source Python package for integrative Raman spectroscopy data analysis. Analytical chemistry, 96(21), 8492-8500.

[**] Pavlou, E., & Kourkoumelis, N. (2025). PyFasma: an open-source, modular Python package for preprocessing and multivariate analysis of Raman spectroscopy data. Analyst, 150(14), 3112-3122.

 

Add to section 6.3 Machine Learning Models for Spectroscopic Classification and after the Elastic Net discussion:

Beyond conventional regularization approaches, specialized spectroscopy-oriented architectures have recently been proposed to preserve the physical interpretability of Raman spectra during model training. Peak Sensitive Elastic Net Regularization (PSE-LR) extends traditional Elastic Net methods by assigning greater importance to diagnostically relevant Raman bands while penalizing model dependence on spectral background fluctuations [***]. By encouraging feature selection around known biochemical signatures, PSE-LR helps maintain the physical integrity of spectral interpretations and reduces the risk of models exploiting noise-driven correlations. Such approaches represent an important step toward briding predictive performance and explainability in clinical spectroscopy.

[***] Wang, Z., Ranasinghe, J. C., Wu, W., Chan, D. C., Gomm, A., Tanzi, R. E., ... & Huang, S. (2025). Machine Learning Interpretation of Optical Spectroscopy Using Peak-Sensitive Logistic Regression. ACS nano, 19(16), 15457-15473.

Add to section 6.3.3 Deep Neural Networks, after the introduction to CNNs and before moving to diagnostic examples.

Recent deep learning developments have produced architectures specifically designed for spectroscopic data rather than adapted from image-analysis frameworks. Examples include Spectra Segmentation Networks (SSNet) [****] and Raman Analysis through Deep Adaptive Representation (RADAR) [*****], which are designed to accomodate the one-dimensional structure of spectral signals while performing automatic feature extraction without manual peak selection. Unlike traditional machine learning pipelines that depend on handcrafted spectral features, these end-to-end architectures learn hierarchical representations directly from raw of minimally processed spectra. By simultaneously identifying diagnostically relevant spectral regions and performing classification, they reduce analyst bias and improve scalability for large clinical datasets.

[****] Yao, M., Zhang, Y., Liu, G., & Pang, D. (2024). SSNet: A novel transformer and CNN hybrid network for remote sensing semantic segmentation. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 17, 3023-3037.

[*****] Sjöberg, J., Siminea, N., Păun, A., Lita, A., Larion, M., & Petre, I. (2025). Radar: raman spectral analysis using deep learning for artifact removal. Advanced Optical Materials, 13(25), 2500736.

 

• Several sections describe numerous studies individually, making it difficult for readers to compare results. The manuscript would benefit from summary tables containing disease, biomarker, plasma/serum sample type, SERS substrate, detection strategy, machine-learning method, sensitivity/specificity/AUC, and limit of detection.

We thank the reviewer for this valuable suggestion. We agree that the original manuscript would benefit from summary tables. We have then included the following:

New Table 1) - Comparative overview of direct (label-free) and indirect (labelled) SERS detection strategies for plasma-based disease diagnostics.

New Table 2) - Comparison of physicochemical and analytical properties of most common SERS-active nanoparticle systems used in plasma-based diagnostics.

New Table 3) - Most common nanoparticle geometries used in SERS-based plasma diag-nostics: structural characteristics, enhancement properties, and analytical relevance.

New Table 4) Comparative overview of biological matrices other than blood plasma or serum, in SERS-based diagnostics: collection characteristics, biomarker content, and matrix-specific analytical considerations.

 

• One of the issues with SERS is mitigating inherent noise when using plasmonic nanoparticles alone. Could the authors elaborate on different pathways to manage the noise in SERS with references (eg – spread spectrum SERS allows label-free detection of attomolar neurotransmitters, Noise management of surface-enhanced Raman spectroscopy using two-dimensional materials, Enhancement of the signal-to-noise ratio in fiber-optics based SERS detection by rough-cutting the end surface).

We thank the reviewer for this suggestion. To address this point, we expanded Section 6.5.3 ("Spectral Instability and Concentration Effects") to discuss recent strategies for noise management in SERS. Specifically, we added discussion and references covering (i) spread-spectrum SERS for computational suppression of spectral noise, (ii) hybrid plasmonic/2D-material substrates for fluorescence quenching and improved SNR, and (iii) optical-engineering approaches such as fibre-optic probe optimisation to reduce instrumental noise. These additions highlight the ongoing shift from maximising enhancement factors towards improving signal-to-noise ratio and reproducibility in clinically relevant SERS systems.

Beyond concentration-dependent variability, SERS measurements are inherently affected by noise arising from stochastic hotspot formation, nanoparticle aggregation, fluorescence background, and local substrate heterogeneity. Consequently, recent research has increasingly focused on improving signal-to-noise ratio (SNR) and measurement reproducibility rather than solely maximizing enhancement factors. Several complementary approaches have emerged to address these limitations. At the computational level, spread-spectrum SERS (SS-SERS) distributes spectral information across a broader frequency domain and reconstructs signals through correlation-based decoding, enabling substantial suppression of uncorrelated noise and fluorescence background while achieving attomolar detection of neurotransmitters [*]. At the substrate level, hybrid plasmonic platforms incorporating two-dimensional materials such as graphene, MoS₂, and WSe₂ have demonstrated reduced spectral variability, fluorescence quenching, and significant SNR improvements compared with conventional metallic nanoparticle substrates [**]. Noise reduction can also be achieved through optimisation of the optical acquisition system. For example, rough-cutting the end surface of fiber-optic SERS probes has been shown to reduce optical artefacts and improve signal collection efficiency, resulting in measurable SNR gains [***]. Collectively, these developments reflect a broader transition from enhancement-driven substrate design towards integrated strategies that simultaneously optimise sensitivity, reproducibility, and noise suppression, which are increasingly recognised as essential requirements for clinical translation of SERS-based diagnostic systems.

[*] Lee W, Kang BH, Yang H, et al. Spread Spectrum SERS Allows Label-Free Detection of Attomolar Neurotransmitters. Nature Communications. 2021;12:159. DOI: 10.1038/s41467-020-20413-8

[**] Ranasinghe JC, Sanders SK. Noise Management of Surface-Enhanced Raman Spectroscopy Using Two-Dimensional Materials. ACS Sensors. 2026;11(3):1920–1932. DOI: 10.1021/acssensors.5c03074

[***] Shin M, Kim K, Jeong DH. Enhancement of the Signal-to-Noise Ratio in Fiber-Optics Based SERS Detection by Rough-Cutting the End Surface. Optics Express. 2023;31(8):12645–12652. DOI: 10.1364/OE.48502

 

• The manuscript repeatedly states that SERS has strong diagnostic potential but has not yet achieved widespread clinical adoption. The authors should provide a better discussion on this. 

We thank the reviewer for their valuable suggestion. We have addressed it briefly in Section 5, mostly as an additional response to reported research performance where suggestions are made on further experimentation, larger clinical cohort validation, and other prerequisite for clinical translation.  Section 7 details the causes/challenges, and opportunities from methods aimed at resolving identified challenges. Most importantly, it is generally implied in the review that once the identified challenges are addressed, and inter-laboratory validation of SOP for substrates, methods, equipment and other variables are created, alongside larger clinical validation from larger cohorts, then SERS would be positioned to be adopted as a viable diagnostic technique.

 

Reviewer 2 Report

Comments and Suggestions for Authors

The submitted manuscript provides a well-structured overview of gold- and silver-nanoparticle-based SERS platforms for clinical plasma diagnostics, effectively highlighting the integration of AI in tackling biofluid analysis. The review accurately highlights the technology's potential and the critical challenges associated with translation, including the need for standardized protocols. Minor revisions are suggested to explicitly define the review's unique value proposition.

Comments.

1. This would be fruitful to dedicate a specific subsection or at least short paragraph to Au/Ag combined with 2D materials (Graphene, MXenes, or MoS₂) and explain how the hybrid substrates provide a dual enhancement mechanism: electromagnetic mechanism from metals and chemical one via charge transfer from the 2D materials. Crucially note that Graphene/MXene encapsulation acts as a protective barrier against silver oxidation and effectively quenches plasma-induced autofluorescence, drastically improving the signal-to-noise ratio in direct SERS.

2. Since the authors brilliantly noted Pseudoreplication in section 6.5.2, this would be nice to provide the community with the explicit methodological cure in the Data Landscape section, e.g. to draw a hard line between Patient-level splitting and Spectrum-level splitting in ML pipelines. Is it possible to explain that taking multiple Raman spectra from a single patient's plasma sample and distributing them across both training and testing folds causes severe Data Leakage?

3. Machine learning models are often dismissed by clinicians as untrustworthy "black boxes." Could the authors detail how Explainable AI (XAI) bridges the gap between deep learning outputs and physical Raman scattering laws?

< !--TgQPHd|[]-->  

Author Response

 

We thank the reviewer for their revision of our manuscript and suggestions given. Following, our point-by-point answers in blue text, with corrections highlighted in yellow in the revised manuscript.

1) This would be fruitful to dedicate a specific subsection or at least short paragraph to Au/Ag combined with 2D materials (Graphene, MXenes, or MoS₂) and explain how the hybrid substrates provide a dual enhancement mechanism: electromagnetic mechanism from metals and chemical one via charge transfer from the 2D materials. Crucially note that Graphene/MXene encapsulation acts as a protective barrier against silver oxidation and effectively quenches plasma-induced autofluorescence, drastically improving the signal-to-noise ratio in direct SERS.

We thank the reviewer for raising this relevant matter, and have addressed in subsection 3.4. Hybrid Nanostructures

Recently developed hybrid architectures composed of noble metal Au/Ag NPs coupled with two-dimensional (2D) materials such as graphene, transition metal carbides/nitrides (MXenes), or transition-metal dichalcogenides like molybdenum disulfide (MoS₂), represent an important advancement in surface-enhanced Raman spectroscopy (SERS), yielding exceptional sensitivity through a synergistic dual enhancement mechanism [80]. The primary contribution arises from the classical electromagnetic mechanism (EM) localized at the noble metal domains. Upon laser excitation, the Au/Ag nanoparticles sustain localized surface plasmon resonance (LSPR), generating intensely amplified localized electromagnetic fields, or "hot spots," at the nanostructured junctions that scale the Raman scattering cross-section by several orders of magnitude [81], [82]. Concurrently, the integrated 2D material sub-layer provides a complementary chemical enhancement mechanism (CM). Owing to their unique electronic band structures and high surface area, materials like graphene, MXenes, or MoS₂ engage in rigorous interfacial molecular interactions with adsorbed analytes. This proximity facilitates efficient photo-induced charge transfer pathways between the highest occupied/lowest unoccupied molecular orbitals (HOMO/LUMO) of the analyte and the Fermi levels of the 2D substrate. This electronic coupling alters the molecular polarizability of the target, effectively mitigating fluorescence background noise and preserving quantitative linearity while working in tandem with the metal-driven EM framework to drastically lower detection limits [81], [82]

2) Since the authors brilliantly noted Pseudoreplication in section 6.5.2, this would be nice to provide the community with the explicit methodological cure in the Data Landscape section, e.g. to draw a hard line between Patient-level splitting and Spectrum-level splitting in ML pipelines. Is it possible to explain that taking multiple Raman spectra from a single patient's plasma sample and distributing them across both training and testing folds causes severe Data Leakage?

We thank the reviewer for this important suggestion. To address this point, we expanded Section 6.4 (Data Landscape in Spectroscopic AI) by explicitly distinguishing spectrum-level and patient-level splitting in SERS-based machine learning workflows. The revised text now explains how distributing multiple SERS spectra from the same patient across training and testing sets introduces severe data leakage, allowing models to learn patient-specific characteristics, substrate-related artefacts, or acquisition signatures rather than disease-associated biochemical features. We further clarify that all spectra from a given individual should remain within a single dataset partition and discuss patient-level splitting as the preferred evaluation strategy. Because limited cohort sizes are common in SERS studies, we also added a brief discussion of appropriate data augmentation practices, emphasizing that augmentation should only be performed after patient-level partitioning and cannot substitute for the biological variability provided by independent patient cohorts. This addition complements the discussion of pseudoreplication in Section 6.5.2 and provides practical guidance for constructing robust SERS-AI pipelines.

Paragraphs added to 6.4. Data Landscape in Spectroscopic AI:

A critical distinction must be maintained between spectrum-level splitting and patient-level splitting. In SERS-based artificial intelligence workflows, dataset integrity is compromised when multiple SERS spectra acquired from the same patient sample are treated as independent observations. If these non-independent spectra are distributed across both training and testing folds, severe data leakage occurs, allowing the model to learn patient-specific characteristics, substrate-related artefacts, or acquisition-specific signatures rather than disease-associated biochemical features. As a result, reported performance metrics may be artificially inflated and fail to reflect true generalisation to unseen patients.

To mitigate this issue, data partitioning should be performed at the patient level, ensuring that all SERS spectra originating from a given individual are assigned exclusively to either the training, validation, or testing set. Patient-level splitting is therefore considered a prerequisite for reliable model evaluation and should be complemented, whenever possible, by external cohort validation to assess robustness across different patient populations, acquisition conditions, and SERS platforms.

Nevertheless, practical limitations such as rare diseases, limited sample availability, and pilot-scale studies often result in small patient cohorts. In these situations, multiple SERS spectra acquired from the same patient may still provide valuable information for model development and data augmentation. However, augmentation procedures should only be applied after patient-level partitioning has been performed, with all original and augmented spectra remaining within the same dataset subset. Strategies such as intensity scaling, spectral perturbation, baseline variation, noise injection, and synthetic spectrum generation can improve model robustness and training stability, but they cannot replace the biological diversity obtained from independent patient cohorts. Consequently, augmentation should be viewed as a complement to, rather than a substitute for, rigorous patient-level validation.

3) Machine learning models are often dismissed by clinicians as untrustworthy "black boxes." Could the authors detail how Explainable AI (XAI) bridges the gap between deep learning outputs and physical Raman scattering laws?

We thank the reviewer for this insightful comment. To address the concern regarding the perceived “black-box” nature of machine learning models, we substantially expanded Section 6.3.6 (Explainable AI and Model Interpretability Modalities). The revised text now explicitly discusses how explainable AI methods, including Grad-CAM, Integrated Gradients, and SHAP, connect model predictions to specific SERS spectral regions and wavenumbers. We further clarify that these attribution methods allow predictions to be interpreted in terms of known vibrational assignments and biochemical signatures, thereby linking deep learning outputs to the physical foundations of Raman scattering and molecular spectroscopy. Examples from the reviewed literature were incorporated to illustrate how XAI can identify diagnostically relevant spectral bands and provide chemically meaningful explanations for model decisions, helping to bridge the gap between predictive performance and clinical trust.

Paragraphs added to 6.3.6. Explainable AI (XAI) and Model Interpretability Modalities:

Explainable AI bridges the gap between deep learning predictions and the physical basis of SERS by transforming abstract model activations into interpretable attribution maps associated with specific spectral regions and wavenumbers [259], [251]. These explanations allow researchers and clinicians to determine whether model predictions are driven by diagnostically relevant vibrational signatures rather than artefacts arising from instrumental variability, background interference, or noise [180], [238]. As a result, XAI provides a mechanism for validating that learned representations remain consistent with established spectroscopic principles and biochemical knowledge.

Added to subsection Gradient-Based Visualization Methods:

By linking model predictions to specific spectral features, XAI enables comparison be-tween algorithmic decision-making and established vibrational spectroscopy assignments [18], [237], [251]. For example, in studies targeting neurodegenerative diseases, attribution maps can identify whether predictions are associated with diagnostically relevant bands such as the 1607 cm⁻¹ uric acid/tryptophan feature or the 1157 cm⁻¹ NeuAc-related vibration [18]. This correspondence provides a physically interpretable rationale for model outputs and strengthens confidence that classification decisions are based on meaningful biochemical information rather than spurious correlations.

Added to subsection Game-Theoretic Feature Attribution (SHAP):

By employing game-theoretic SHAP values, each diagnostic prediction can be decomposed into contributions from individual spectral features. This enables quantification of how specific wavenumber regions influence the probability assigned to a disease class, providing a transparent connection between model outputs and the molecular vibrations represented within the spectrum [263], [264], [237]. Consequently, SHAP-based explanations help ensure that predictive performance can be interpreted in the context of underlying biochemical and spectroscopic mechanisms.

 

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript ID sensors-4371325 has been devoted to mainly present a review about gold- or silver-nanoparticle surface-enhanced Raman spectroscopy systems. The authors focused on the study of potential applications for plasma-based diagnostics and AI-driven analysis. Please see below a list of comments to the authors:

1. The authors should clearly state in the introduction what this work adds to literature in respect to previous publications in the topic of nanoparticle surface-enhanced Raman spectroscopy systems.

2. The abstract is missing.

3. A roadmap with the progress of the topic of nanoparticle surface-enhanced Raman spectroscopy would be welcome.

4. A graphical representation to show the aim of the work, the parameters analyzed and the main findings could be incorporated in the introduction section.

5. Advantages of cutting edge surface-enhanced Raman spectroscopy systems could be summarized in section 7.

6. Please comment about the importance of the size and isotropy or anisotropy of the nanoparticles for SERS systems.

7. Currently, how is the dependence on incident polarization of the nanoparticles proposed for SERS enhancement?

8. The authors are invited to separate the advantages of hybrid or hierarchical nanostructures based on metallic nanoparticles, please see: https://doi.org/10.3390/bios14020108

9. Some comments about structured light nanoparticles-SERS can be considered to improve the challenges, you can see for instance:

https://doi.org/10.1021/acsphotonics.5c01391

10. In my opinion, some references cited in collective form could be split in order to better justify the importance of the individual information related to each citation selected for the analysis of the topic.

Comments on the Quality of English Language

A proofreading is suggested

Author Response

We thank the reviewer for their revision of our manuscript and suggestions given. Following, our point-by-point answers in blue text, with corrections highlighted in green in the revised manuscript.

1) The authors should clearly state in the introduction what this work adds to literature in respect to previous publications in the topic of nanoparticle surface-enhanced Raman spectroscopy systems.

We thank the reviewer for their valuable suggestion. We have expanded the introductory paragraphs in Section 1 to explicitly articulate the unique scope of this review, namely, the convergence of blood plasma as the sole biological matrix, AuNP/AgNP as the plasmonic platforms, and AI-driven spectral analysis. Distinguishing it from prior SERS reviews that address broader matrices or omit systematic chemometric discussion. This one was also of the suggestions of reviewer 1, so the added text to Section 1 is in blue.

2) The abstract is missing.

We thank the reviewer for noticing this. In fact, we did submit an Abstract online, that I reproduce below for reviewer inspection, but the instructions for authors did not mention an Abstract in Review article:

“Surface-enhanced Raman spectroscopy (SERS) has emerged as a highly promising analytical technique for disease diagnostics due to its exceptional sensitivity, molecular specificity, and ability to detect a broad range of biomarkers in complex biological matrices. This review provides a comprehensive overview of gold- and silver-nanoparticle-based SERS platforms for plasma disease diagnostics, covering advances in plasmonic nanostructures, biological sample analysis, biomarker detection, and AI-driven spectral data processing. Particular emphasis is placed on the application of SERS to clinically relevant biofluids, especially plasma, where the technique has demonstrated considerable potential for detecting diseases such as cancer, inflammatory disorders, and neurological conditions. The review also critically examines the major challenges currently limiting the clinical translation of SERS technologies. These include variability associated with substrate fabrication, matrix-induced signal fluctuations, limited interlaboratory reproducibility, and the lack of standardized protocols for spectral preprocessing and data analysis. Strategies proposed to address these issues are discussed, including comprehensive post-synthesis substrate characterization, optimization of biological sample preparation, advanced spectral preprocessing workflows, and the integration of machine learning and artificial intelligence algorithms to improve diagnostic robustness and reproducibility. Collectively, the advances summarized in this review indicate that SERS-based diagnostic technologies are rapidly progressing beyond proof-of-concept studies toward clinically applicable systems. Continued interdisciplinary collaboration and standardization efforts will be essential to bridge the remaining gap between experimental SERS methodologies and routine clinical implementation.”

3) A roadmap with the progress of the topic of nanoparticle surface-enhanced Raman spectroscopy would be welcome.

We appreciate the reviewer's suggestion and recognize the scientific curiosity that motivated it. However, a thorough discussion of this broad and complex topic falls outside the scope of the present study. Therefore, we refer interested readers to dedicated review articles that address these issues in greater depth, such as, for example, Langer et al., ‘Present and Future of Surface-Enhanced Raman Scattering’, ACS Nano, vol. 14, no. 1, pp. 28–117, Jan. 2020, doi: 10.1021/acsnano.9b04224; or Yi et al., ‘Surface-enhanced Raman spectroscopy: a half-century historical perspective’, Chem. Soc. Rev., vol. 54, no. 3, pp. 1453–1551, 2025, doi: 10.1039/D4CS00883A.

4) A graphical representation to show the aim of the work, the parameters analyzed and the main findings could be incorporated in the introduction section.

We thank the reviewer for this valuable suggestion. Accordingly, the revised manuscript now includes Figure 1, which provides an overview of the aim, objectives, and key findings emerging from the literature reviewed.

Figure 1. Schematic overview of the scope and main findings of the literature reviewed. (A) SERS fundamentals, enhancement mechanisms, and nanoparticle platforms used for biosensing applications. (B) Biomolecules contributing to plasma-derived SERS signatures and their relevance to disease diagnosis. (C) AI-driven models and analytical workflows for the processing, classification, and interpretation of SERS spectral data.

5) Advantages of cutting edge surface-enhanced Raman spectroscopy systems could be summarized in section 7,.

We thank the reviewer for this valuable suggestion. Accordingly, we have addressed it in section 7.1.5. Opportunities:

The significant improvement in component power, quality, and equipment performance, accompanied by lower price and compact size improves the feasibility of SERS for general and POC deployment [62]. Inasmuch as SERS is effective as a standalone technique, developing multimodal Raman spectrometers integrated with efficient sample handling techniques can enhance automation and protocol standardisation [115]. Beyond hardware improvements, advanced software and computational approaches can help overcome challenges associated with complex SERS data analysis and improve diagnostic accuracy. In particular, integrating data-driven and knowledge-based algorithms may enhance both the interpretability and reproducibility of SERS results [43], [111].

Within the SERS research landscape, cutting-edge advances are being applied to resolve the identified challenges. For example, microfluidic integration represents an opportunity that simultaneously addresses matrix complexity, measurement reproducibility, and point-of-care applications [263]. Platforms such as the cardiac troponin I detection system reported by Campu et al. [97] and the acoustofluidic multimodal system for Alzheimer's biomarker isolation developed by Hao et al. [102] demonstrate that microfluidic sample handling can reduce protein corona formation, control analyte concentration at the SERS-active surface, and enable automated sequential detection. Proceeding without requiring skilled operator intervention. The extension of such platforms to include on-chip sample pretreatment (dilution, pH adjustment, filtration) would further reduce pre-analytical variability and bring SERS-based plasma diagnostics within reach of near-patient or point-of-care deployment. The fabrication costs of microfluidic SERS chips have decreased substantially with advances in soft lithography and injection moulding, and their scalability for clinical-volume production is no longer a prohibitive constraint. Also, microfluidic configurations with multiplex capability such as spatial multiplex, barcode multiplex and Label-free multiplex are significant and have the potential to simultaneously detect a panel of disease biomarkers [263]. Cutting-edge fibre optic technology is being adapted for biosensing applications, where readily functionalised fibre surfaces can be integrated with SERS to provide highly selective light-matter interaction depending on various transduction mechanisms.

Multimodal detection strategies offer an empirical route to minimise the false-positive and false-negative rates that limit single-mode SERS platforms. As demonstrated in Section 5, combining SERS with complementary transduction mechanisms like electrochemical, colorimetric, photothermal, or acoustic, enables orthogonal result validation within a single measurement workflow. It also reduces the impact of substrate-induced signal variability on diagnostic conclusions, and can access biomarker information that is not spectrally accessible by SERS alone. The development of multimodal platforms is not restricted to highly sophisticated laboratory systems, as exemplified by the SERS–colorimetry lateral flow strip reported by Shende et al. [264] for simultaneous codeine and fentanyl detection in plasma without sample pretreatment. Simple dual-mode confirmation can be incorporated into formats compatible with decentralised testing.

AI-driven data analysis with clinical-grade validation presents an immediate opportunity to improve the translational credibility of existing SERS datasets. Many of the plasma studies reviewed in Section 5 contain spectral datasets of sufficient depth to support rigorous validation frameworks extended to real patients/samples cross-validation, large external cohort testing, and explainability mapping. Retrospective analyses of archived SERS datasets is capable of creating a comparative benchmark (currently unavailable) that can be used to determine viable clinical evaluation. Remarkably, the array of AI-based algorithms discussed in Section 6, with many models tailored to spectroscopic data analyses for example PSE-LR, RamanSpy, PyFasma, SSNet, RADAR are helping make SERS more quantitative.

6) Please comment about the importance of the size and isotropy or anisotropy of the nanoparticles for SERS systems.

We thank the reviewer for raising this very relevant matter, and have further addressed it in section 3.1. Properties of SERS-Active Nanoparticles. We have also added a convenient new Table 3) - Most common nanoparticle geometries used in SERS-based plasma diagnostics: structural characteristics, enhancement properties, and analytical relevance.

The diverse applications of noble metal NPs stem from unique optical, magnetic, catalytic, and therapeutic properties that emerge under nanoscale spatial confinement, where electronic structures transition from semicontinuous to discrete. Specifically, gold nanoparticles with a core diameter of 3 nm or larger adopt a face-centered-cubic (FCC) structural framework with a semicontinuous electronic profile, enabling coherent collective electronic oscillations known as surface plasmon resonance [69]. Conversely, when dimensions fall below the 3 nm threshold, the particles cannot support plasmon resonance; their resulting discrete electronic states trigger molecular-like properties, classifying them as nanoclusters. These distinct electronic configurations, combined with exceptionally high surface-to-volume ratios, render the physicochemical properties of noble metal nanoparticles highly sensitive to absolute size, elemental composition, intraparticle atomic packing, and interparticle hierarchical arrangement. Consequently, this profound sensitivity has driven the development of advanced, precise synthetic methodologies to control these property-dictating attribute s[69].

The SERS optimized dimensions across nanostructures follow a clear size–function relationship [69], [70]. Nanospheres perform best with core diameters between 30–80 nm, a range that preserves strong dipolar plasmon modes while avoiding both quantum damping at small sizes and multipolar scattering at larger ones. Nanorods require more than one dimensional constraint: their 20–40 nm core diameters maintain a clean longitudinal dipole resonance, 50–150 nm lengths tune the LSPR to match the excitation wavelength, and <5–15 nm tip radii create intense localized hotspots. Nanoprisms rely on thin 10–30 nm cores to support strong in plane dipole modes, 50–150 nm edge lengths to set the resonance wavelength, and <5–10 nm tip radii to maximize field enhancement at their vertices. Finally, nanostars, whose performance is dominated by curvature, achieve their strongest hotspots when their tips are sharpened to <2–10 nm, enabling extreme electromagnetic localization [69], [70]. What is obvious is that NP size determines its light-absorption and scattering cross-sections and the efficiency of LSPR. Geometric isotropy or anisotropy determines the spatial distribution of plasmonic energy, with anisotropic shapes concentrating the electromagnetic field into highly intense "hot spots" at sharp vertices rather than distributing it uniformly like isotropic spheres

7) Currently, how is the dependence on incident polarization of the nanoparticles proposed for SERS enhancement?

We thank the reviewer for raising this important matter. The answer is addressed in answer 9.

8) The authors are invited to separate the advantages of hybrid or hierarchical nanostructures based on metallic nanoparticles, please see: https://doi.org/10.3390/bios14020108

We thank the reviewer for raising this relevant matter, and have addressed in subsection 3.4. Hybrid Nanostructures

A different class of hybrid nanomaterials are hierarchical nanostructures possessing large reaction interface in select surface area allowing for superior biomolecular detection, catalyst charge transfer, metal ion release, and cavities effective for microbial capture. Their adjustable porosity, packing density, and controlled stability makes them suitable for biosensing applications [79]. With advantages for SERS assays due to ultra-high sensitivity and low detection limits through pure EM amplification, in addition to large area uniformity, excellent reproducibility, and practical reusability.

9) Some comments about structured light nanoparticles-SERS can be considered to improve the challenges, you can see for instance: https://doi.org/10.1021/acsphotonics.5c01391

We thank the reviewer for raising this important matter, and will answer it as a joint challenge with the polarization issue raised in question 7. The following concise paragraph was included in the “Challenges” section, subsection 7.1.4. Experimental and Equipment Challenges:

Another challenge is the polarization sensitivity of plasmonic nanoparticles, as SERS enhancement depends not only on nanoparticle geometry but also on the incident polarization and structured-light excitation used. Variations in near-field distribution and Raman tensor coupling can significantly affect signal intensity and spectral response, complicating the standardization of AI-assisted plasma diagnostic platforms [277].

10) In my opinion, some references cited in collective form could be split in order to better justify the importance of the individual information related to each citation selected for the analysis of the topic.

We thank the reviewer for this perfectly logical suggestion, and have made an effort, all along the ms to split references as much as possible in order to better justify the importance of individual information.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript has been sufficiently improved to warrant publication in Sensors. I approve the revised version for publication.

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have clarified all the points raised in the initial review stage. The paper can be a base for future research and then I can recommend it for publication in present form.

Comments on the Quality of English Language

A proofreading is suggested