Sage Journals: Discover world-class research

Abstract

The accurate detection of pesticide residues in fresh produce is essential for public health. This study evaluated the use of two-dimensional correlation spectroscopy (2D-COS), specifically applied to Raman data, combined with principal component analysis (PCA) and support vector machines (SVMs) for the direct and rapid detection of chlorothalonil pesticide residues in four different vegetable matrices. This approach significantly enhances the spectral resolution, enabling the identification of subtle chlorothalonil-specific fingerprints. Raman spectra were analyzed across the full wave range (300–2500 cm^–1), with four key fingerprint regions identified through 2D-COS: 354−414 cm^–1 (C–Cl stretching), 1260−1286 cm^–1 (C–C stretching), 1540−1570 cm^–1 (C–C stretching), and 2250−2265 cm^–1 (C≡N stretching). When applied to the full spectral range, the PCA-SVM model yielded only moderate classification accuracy (accuracy = 72%, κ = 0.43). In contrast, models focused on the fingerprint regions achieved perfect classification (accuracy = 100%, κ = 1), successfully distinguishing between the spiked and control samples. The consistent detection of C–Cl and C≡N bonds in all vegetable matrices highlights their reliability as universal markers for chlorothalonil detection. These findings demonstrate the potential of Raman spectroscopy combined with targeted data analysis, as a highly sensitive and reliable method for pesticide residue screening.

Graphical abstract

This is a visual representation of the abstract.

Keywords

Pesticide residues Raman spectroscopy two-dimensional correlation spectroscopy 2D-COS chlorothalonil molecular fingerprints machine learning principal component analysis PCA support vector machine SVM

Introduction

The widespread and often indiscriminate use of synthetic pesticides in modern agriculture poses significant public health risks due to residue contamination.^1–3 Organochlorine pesticides (OCPs), in particular, have been linked to a range of health issues, including reproductive, immune, and neurological damage, as well as an increased risk of cancer.^4,5 Although many developed countries have banned OCPs, they remain in use in many developing countries.^6,7 This ongoing use highlights an urgent need for methods capable of rapidly screening fresh produce and environmental samples to ensure safety and regulatory compliance, especially in areas where these pesticides are still applied. Conventional methods for pesticide residue analysis, such as gas chromatography–mass spectrometry and high-performance liquid chromatography, are highly sensitive and accurate but are also labor-intensive, time-consuming, and require complex sample preparation.^8,9 In contrast, vibrational spectroscopy techniques such as near-infrared (NIR)¹⁰ and Raman spectroscopy¹¹ have emerged as promising alternatives for pesticide residue analysis.^12–15 These techniques offer several advantages, including rapid analysis, portability, low operational costs, and ability to conduct direct, non-destructive testing on samples, making them particularly suitable for targeted screening of large volumes of fresh produce.

Identifying unique vibrational spectral markers (also referred to as fingerprints) that indicate specific pesticide residues is often complicated by matrix effects, spectral overlap, and noise. To address this, multivariate data analysis techniques such as partial least squares discriminant analysis,^16,17 multivariate curve resolution,^18,19 and analysis of variance²⁰ have been applied. Although these methods assist in resolving vibrational spectra, overlapping bands and noise continue to present interpretability challenges. Another approach involves modeling theoretical vibrational spectra using density functional theory (DFT) calculations, which have been widely used to simulate the Raman spectra of pesticides.^21–25 However, discrepancies between theoretical and experimental spectra, often caused by variations in molecular conformation and interactions, limit the broader applicability of DFT in fingerprint identification. Additionally, the high computational cost of DFT, coupled with the need for costly analytical-grade standards, further restricts its widespread application.

Two-dimensional correlation spectroscopy (2D-COS) is a powerful technique used to analyze system responses to external perturbations.^26–32 It has been effectively applied across various spectroscopic methods, including infrared (IR), NIR, Raman, nuclear magnetic resonance, fluorescence, and ultraviolet–visible, providing enhanced resolution of overlapping spectral features.^29,33 This capability makes 2D-COS particularly useful for detecting subtle spectral changes and identifying correlations that are often obscured in conventional Raman spectra. In this study, we explore, for the first time, the potential of 2D-COS to identify unique Raman fingerprints indicative of pesticide residues, specifically chlorothalonil (an OCP), across various vegetable matrices, including kale, pigweed, tomato, and bell pepper. We then apply principal component analysis (PCA) and support vector machines (SVMs) to develop classification models for targeted detection. Previous studies using Raman spectroscopy for chlorothalonil detection have relied on surface-enhanced Raman spectroscopy (SERS), focusing on a single Raman fingerprint and one sample matrix.^34,35 Although SERS provides high sensitivity, it is relatively costly compared to conventional Raman combined with 2D-COS due to the high cost of nanoparticle synthesis and sample preparation, making it less feasible for routine, field-based applications, especially in resource-limited settings.

Experimental

Materials and Methods

Sample Preparation, Data Acquisition, and Pre-Processing

Two standard solutions of chlorothalonil pesticide were prepared at different concentrations: (i) a control solution containing only deionized water and (ii) a spiked solution with a concentration of 2 g/L of chlorothalonil, prepared by diluting a commercially available formulation (480 g/L chlorothalonil) with deionized water as recommended by the label instructions. Fresh, organically grown bell peppers (Capsicum annuum), kale (Brassica oleracea var. sabellica), tomatoes (Solanum lycopersicum), and pigweed (Amaranthus spp.), free of blemishes, were selected for analysis. The vegetables were thoroughly cleaned with deionized water and allowed to dry for 15 min to remove surface contaminants. Each vegetable sample was then randomly divided into two groups: one group was sprayed with the control solution, and the other with the spiked solution. After application, the vegetables were left to dry completely for 30 min to ensure full solvent evaporation and proper adherence to the pesticide residue before spectral collection.

Raman spectra were collected using a 785 nm excitation source (EZ Raman Spectrometer, Enwave Optronics). The spectrometer was calibrated using a polystyrene standard. To ensure reproducibility and account for matrix inhomogeneity, 20 spectra were collected from five different locations on each vegetable sample. The laser spot size was calculated to be 4.35 µm based on the 100 µm fiber core diameter and a numerical aperture of 0.22, providing a balance between spatial resolution and signal intensity. Optimal acquisition parameters, i.e., 200 mW laser power and a 5 s exposure time, were selected for maximizing signal intensity while minimizing noise. The fresh vegetable samples showed no visible signs of damage during measurements. Spectra were limited to the 300−2500 cm^–1 range and the vector normalized between 0 and 1 to standardize intensity values for comparison across samples.

Two-Dimensional Correlation Spectroscopy (2D-COS) Analysis and Machine Learning Model Development

For 2D-COS, a mean spectrum for each class of the Raman spectra served as a reference for the Fourier transform correlations. As a result, four distinct correlation spectra were obtained from the four different vegetable matrices. The 2D-COS Raman spectra were generated by plotting 40 equally spaced contour levels with intensity values ranging from 8 to 19, resulting in a clean contour plot of the correlation spectrum. This approach highlights the most significant changes while excluding very low or very high intensities that may not be of interest or might clutter the plot. Following this, PCA³⁶ was conducted on the identified fingerprint bands (regions of high perturbations) from the 2D-COS analysis to reduce dimensionality and visualize the data. Subsequently, the principal components accounting for the most significant cumulative variance were extracted (two for each case) and utilized to develop a linear SVM³⁷ model to classify the control and pesticide-spiked samples. All analyses were carried out in open source R v.3.6.3 using Corr2D for the 2D-COS analysis,³³ ChemoSpec for PCA,³⁸ and e1071 for SVM.³⁹ Additionally, we compare the performance of 2D-COS analysis with difference spectroscopy whose approach involves subtracting spectra to remove common features thereby highlighting subtle technique that highlights subtle changes.⁴⁰

Results and Discussion

Raman Spectral Analysis of Vegetable Samples Highlighting Bio-Molecular and Chlorothalonil-Specific Vibrational Peaks

Figure 1 presents the representative average Raman scattering spectra of four vegetable samples (kale, pigweed, tomato, and bell pepper). Despite being a weak Raman scatterer, the vegetable matrix predominantly absorbs light at low wavenumber regions (below 1500 cm^–1), resulting in a high concentration of peaks in that range. The spectra of the control samples unveiled characteristic peaks corresponding to the bio-molecules commonly found in plant cell walls. Specific bands around 746, 914, 1047, 1268, 1326, 1385, and 1438 cm^–1 indicated the presence of pectin, cellulose, phenylpropanoids, and aliphatic chains. In addition, peaks at 1000, 1153, 1186, 1213, and 1523 cm^–1 suggested the presence of carotenoids, the colored pigments in the samples. The peak at 1690 cm^–1 corresponded to the carbonyl group (C=O) stretching vibration, likely from carboxylic acids. The spectra of the spiked samples exhibited all the peaks observed in the control samples, in addition to extra bands characteristic of the spiking agent, chlorothalonil (C₈N₂Cl₄). The presence of peaks at 326 and 384 cm^–1 confirmed the presence of C−Cl stretching. The 464 and 1252 cm^–1 bands indicated C–C stretching, potentially from the aromatic ring structure within the pesticide. Although the C–C stretching vibration of the carotenoids overlapped with the C–C stretching of the pesticide at 1523 cm^–1, the presence of a distinct peak at 2238 cm^–1 corresponds to the C≡N stretching vibration definitively identified as chlorothalonil in the sample.. A summary of the vibrational bands and their assignments for the vegetable plants is listed in Table I.

Figure 1.

Representative average Raman spectra of the spiked and controlled samples (kale, pigweed, tomato, and bell pepper).

Table 1.

Main Raman vibrational bands identified and their corresponding assignments.

Band (cm⁻¹)	Vibration mode	Assignment	Reference
326	C–Cl stretching	Chlorothalonil (spiked samples only)	⁴¹
384	C–Cl stretching	Chlorothalonil (spiked samples only)	⁴²
464	C–C stretching	Chlorothalonil (spiked samples only)	^21,42
746	Out-of-plane bending of the C–O–H group in the carboxyl group (COOH)	Pectin, chlorophyll A, Carbohydrates	^43,44
914	Stretching (C–O–C) in the plane, symmetric	Cellulose, chlorophyll A phenylpropanoids	⁴⁴
1000	In-plane CH– rocking	Carotenoid	^43,45–49
1047	C–O and C–C stretching, C–O–H bending	Cellulose, phenylpropanoids	⁴⁴
1085	C–O–H bending next to aromatic ring + C–H bending	Carbohydrate	⁴³
1153	C–C stretching; C–O–C bending; C–C stretching in glycosidic linkages; asymmetric ring breathing.	Carotenoid	^{43,45–47,49}
1186	C–O–H bending next to aromatic ring + CH stretching	Carotenoids, phenylpropanoids	^50,51
1213	C–C–H bending	Carotenoids, aliphatics	^50,52
1268	Guaiacyl ring breathing, C–O stretching (aromatic)	Phenylpropanoids	⁴⁴
1285	C–C–H bending	Aliphatics	^44,52
1326	CH₂ bending	Aliphatics, cellulose, lignin	⁴⁴
1385	CH₂ bending	Aliphatics, cuticular waxes	⁴⁴
1438	CH₂+ CH₃ bending	Aliphatics, cutin, cuticular waxes	⁴⁴
1523	−C=C– (in-plane)	Carotenoid	^44,49,53
1603	C–C stretching of aromatic ring + CH stretching	Phenylpropanoids	^44,54,55
1690	C=O stretching	Carboxyl groups	⁴³
2238	C≡N stretching	Chlorothalonil (spiked samples only)	^21,35

Identifying Chlorothalonil Fingerprints and Spectral Variations in Raman Spectra

Difference Spectroscopy Results

Figure 2 presents the difference spectroscopy results, highlighting spectral variations between spiked and control samples of tomato, bell pepper, pigweed, and kale. Significant changes are observed at wave–numbers 342, 394, 445, 470, 531, 584, 611, 722, 750, 913, 1041, 1053, 1146, 1168, 1194, 1270, 1333, 1371, 1516, 1534, 1547, and 2238 cm^–1 corresponding to vibrational modes of chlorothalonil as well as natural components of the vegetable matrices. Despite these observed differences, distinguishing chlorothalonil-specific peaks remains challenging due to overlap with the vegetable matrix and noise amplification. For instance, the C–Cl stretch at 394 cm^–1 and the C≡N stretch at 2238 cm^–1 serve as clear markers of chlorothalonil, yet peaks at 722, 913, and 1270 cm^–1 overlap with vibrational modes from carotenoids, cellulose, and phenylpropanoids. This overlap, particularly within the 500–1500 cm^–1 range, complicates the clear identification of pesticide signals amid the vegetable components. Furthermore, noise amplification during the subtraction process introduces baseline distortions, especially in the 500–1700 cm^–1 region, further hindering the clarity of the spectral data.

Figure 2.

Difference Raman spectra of four vegetable samples (tomatoes, bell pepper, pigweed, and kale). The spectra represent the difference between spiked and control samples, highlighting variations across the spectral range of 300–2500 cm^–1.

Enhanced Resolution of Chlorothalonil Fingerprints Using 2D-COS

To overcome the limitations of difference spectroscopy, 2D-COS was applied to enhance spectral resolution and resolve overlapping bands in the Raman spectra. The synchronous 2D-COS results, shown in Figures 3a–3d, reveal significant autopeaks at 385 cm^–1 (C–Cl stretching), 1263 cm^–1 (C–C stretching), 1548 cm^–1 (C–C stretching), and 2238 cm^–1 (C≡N stretching). These well-resolved peaks across all tested vegetable samples (tomato, bell pepper, pigweed, and kale) provide strong evidence of chlorothalonil contamination. Positive cross-peak correlations indicate molecular interactions between the pesticide and the vegetable matrix, further validating these assignments. The C≡N stretching peak at 2238 cm^–1, a recognized marker for chlorothalonil, stands out as a key indicator of its presence. Additionally, vibrational modes such as the C–Cl stretch at 385 cm^–1 and the C–C stretch at 1263 cm^–1 are clearly distinguished and highly identifiable. The key fingerprint regions identified for chlorothalonil detection through 2D-COS are summarized in Table II.

Figure 3.

Synchronous 2D Raman correlation spectra comparing spiked and control vegetable samples across the wavenumber range of 300 to 2500 cm^–1. The spectra are presented for four different vegetables: (a) tomato, (b) kale, (c) bell pepper, and (d) pigweed. The reference Raman spectra for the spiked and control samples are displayed along the top and left margins, respectively. Positive correlations are indicated in red, and the dashed blue diagonal line highlights the autopeaks common to chlorothalonil at 385, 1263, 1548, and 2238 cm^–1. The 2D Raman correlation spectra were plotted using only positive correlations, focusing on high-intensity data to emphasize the most significant features while excluding low or extreme intensities that could clutter the visualization.

Table II.

Identified fingerprint regions of chlorothalonil from 2D-COS with the highest intensity.

Peak maximum position (cm^–1)	Peak width for PCA (cm^–1)	Attribution
385	354–414	C−Cl stretching
1263	1260–1286	C−C stretching
1548	1540–1570	C−C stretching
2238	2265–220	C≡N stretching

The Raman spectrum of pure commercial-grade chlorothalonil, shown in Figure 4, validates the fingerprints identified by 2D-COS. Major peaks at 387, 1261, 1546, and 2239 cm^–1 confirm chlorothalonil-related vibrations, verifying these regions as reliable markers for pesticide detection in vegetable samples.

Figure 4.

Raman spectra of the pure spiking solution of commercial-grade chlorothalonil, show major peaks at 387, 1261, 1546, and 2239 cm^–1. These peaks validate that the elucidated fingerprints in 2D-COS are attributable to the pesticide, corroborating the regions of interest identified in the vegetable samples.

Principal Component Analysis (PCA)-SVM Results for Classifying Spiked and Control Samples

Principal Component Analysis (PCA)-SVM Results for the Full Spectral Range

To establish a baseline for comparison, PCA was first applied to the full Raman spectrum (300–2500 cm^–1) to assess whether analyzing the entire spectral range could effectively distinguish between spiked and control samples. The results, however, indicated a significant overlap between the spiked and control samples. The PCA score plot (Figure 5a) showed that the first two principal components, PC1 and PC2, explained 43% and 34% of the variance, respectively, accounting for a total variance of 77%. This relatively low explained variance, combined with the dispersed distribution of points, suggested that the full spectrum contains too much irrelevant or overlapping information to allow clear differentiation between the spiked and control groups. The complexity of the vegetable matrix and the overlapping spectral features from both chlorothalonil and the natural vegetable components contributed to this issue. When the full-spectrum PCA results (two principal components that explained the highest cumulative variance) were fed into the SVM model, the classification accuracy was just 71.5%, with a kappa (κ) statistic of 0.43, indicating a high chance of misclassification (Figure 5b). This outcome reflects the limitations of using the entire spectral range, as the noise and spectral overlap reduced the model's ability to reliably distinguish between spiked and control samples.

Figure 5.

(a) Principal component analysis (PCA) of the whole spectra region (300−2500 cm^–1) shows no separation of the spiked and control samples. (b) Scatter plots showing the linear PCA-SVM classification results for the whole region of the spectra (300−2500 cm^–1). The model had a poor classification accuracy of 71.5% and a κ statistic of 0.43. The gray areas indicate the regions where the SVM model predicts the samples to be spiked, while the white areas indicate the regions where the model predicts the control samples.

Principal Component Analysis (PCA)-SVM Results Using Identified Fingerprints

Given the poor performance of the full spectral range analysis, we focused the PCA on specific fingerprint regions known to be relevant to chlorothalonil, i.e., the regions 354−414 cm^–1 (C–Cl stretching), 1260−1286 cm^–1 (C–C stretching), 1540−1570 cm^–1 (C–C stretching), and 2250−2265 cm^–1 (C≡N stretching) correspond to vibrational modes specific to the pesticide and were expected to improve classification accuracy. The results confirmed this expectation, with PCA applied to these regions yielding distinct clusters for spiked and control samples. The cumulative variance explained by the first two principal components was significantly higher for each targeted region: 88%, 96.7%, 92%, and 94.8%, respectively (Figures 6a–6d). The clear separation observed in PCA plots, regardless of vegetable matrix, underscored the ability of these regions to distinctly differentiate between spiked and control samples.

Figure 6.

Score plots of the PCA for the regions from 2D-COS. (a) 354–414 cm^–1, (b) 1260–1286 cm^–1, (c) 1540–1570 cm^–1, and (d) 2250–2265 cm^–1. The plots for all four chlorothalonil fingerprints unequivocally demonstrate the ability to discriminate between control and spiked samples along PC1. The method's effectiveness is underscored by the clear separation of all the samples into two distinct clusters, control (green) and spiked (red), irrespective of the different vegetable matrices.

Furthermore, applying the first two principal components from these fingerprint regions to the SVM model resulted in perfect classification accuracy, achieving 100% with a κ statistic of 1 (Figures 7a–7d). This result highlights the importance of focusing on specific spectral regions that are directly related to chlorothalonil, rather than relying on the entire spectrum, which contains unnecessary and overlapping information. By concentrating on these fingerprint regions, the PCA-SVM model achieved robust and reliable classification, making it a superior method for detecting chlorothalonil residues in vegetables. The consistent detection of C–Cl vibrations (354−414 cm^–1) and C≡N stretching (2238 cm^–1) across all vegetable matrices positions these as unique markers for chlorothalonil contamination, as they are naturally absent in the vegetable matrix. In contrast, C–C bonds in the 1260−1286 cm^–1 and 1540−1570 cm^–1 regions overlap with natural vibrations from the vegetables, but the overall approach still allows for successful differentiation.

Figure 7.

Scatter plots showing the linear PCA-SVM classification results for different spectral regions: (a) 354–414 cm^–1, (b) 1260–1286 cm^–1, (c) 1540–1570 cm^–1, and (d) 2250–2265 cm^–1. Each region achieved a classification accuracy of 100% and a κ coefficient of 1. The gray areas indicate the regions where the SVM model predicts the samples to be spiked, while the white areas indicate the regions where the model predicts the control samples.

Conclusion

This study effectively demonstrates the utility of 2D-COS applied to Raman data, in combination with PCA and SVM, for detecting chlorothalonil residues in various vegetable matrices. While difference spectroscopy highlighted general spectral variations between spiked and control samples, it was insufficient in resolving overlapping signals, especially in complex matrices. 2D–COS offered a more advanced approach, enhancing spectral resolution and enabling the identification of critical C–Cl and C≡N bond vibrations, which served as reliable fingerprints for chlorothalonil across all the vegetable sample types. The perfect classification accuracy achieved by focusing on these fingerprint regions, as opposed to the full spectral range, highlights the value of targeting key vibrational bands. Although this study used higher pesticide concentrations as a proof of concept, the findings lay a solid groundwork for future research at lower concentrations, closer to regulatory maximum residue limits. This methodology shows substantial promise for field-based pesticide screening, offering a rapid, non-invasive, and accurate approach to food safety. Future studies should validate these findings under real-world conditions and examine their effectiveness at detecting lower pesticide residues, moving toward practical, routine food safety monitoring.

Footnotes

Acknowledgments

The authors acknowledge the Center for High-Performance Computing (CHPC), South Africa, for providing computational resources to this research project.

Author Contributions

C. N. Ndung’u: Conceptualization, methodology, investigation, data curation, formal analysis, writing original draft, writing review, and editing. K. A. Kaduki: Conceptualization, funding acquisition, supervision, writing review, and editing. M. I. Kaniu: Conceptualization, supervision, resources. software, methodology, writing review, and editing. L.W. Kiruri: Conceptualization, supervision, methodology, writing review, and editing.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Declaration of Generative Artificial Intelligence (AI) and AI-Assisted Technologies in the Writing Process

During the preparation of this manuscript, the authors utilized an LLM for copy editing and language refinement. The AI tool was employed solely to enhance text clarity and fluency. All content, including citations and references, was independently generated and verified by the authors prior to AI usage. The authors take full responsibility for the accuracy and integrity of the manuscript. The limitations of AI, including potential biases and errors, were acknowledged, and steps were taken to mitigate these in the final manuscript.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Swedish International Development Cooperation Agency (SIDA), through the International Science Programme (ISP), Uppsala University, for financial support through grant numbers IPPS KEN:04 and IPPS AFR:04/AFSIN, and The Kenya Education Network financially supported this project through SIG grant number KENET/CMMS/2022.

ORCID iDs

Charles N. Ndung’u

Kenneth A. Kaduki

References

Leskovac

Petrović

. “Pesticide Use and Degradation Strategies: Food Safety, Challenges, and Perspectives”. Foods. 2023. 12(14): 2709. https://doi.org/10.3390/foods12142709

Mandal

Poi

Hazra

D.K.

Ansary

, et al. “Review of Extraction and Detection Techniques for the Analysis of Pesticide Residues in Fruits to Evaluate Food Safety and Make Legislative Decisions: Challenges and Anticipations”. J. Chromatogr. B. 2023. 1215: 123587. https://doi.org/10.1016/j.jchromb.2022.123587

Onyeaka

Ghosh

Obileke

Miri

, et al. “Preventing Chemical Contaminants in Food: Challenges and Prospects for Safe and Sustainable Food Production”. Food Control. 2024. 155: 110040. https://doi.org/10.1016/j.foodcont.2023.110040

Androutsopoulos

V.P.

Hernandez

A.F.

Liesivuori

Tsatsakis

A.M.

. “A Mechanistic Overview of Health Associated Effects of Low Levels of Organochlorine and Organophosphorous Pesticides”. Toxicology. 2013. 307: 89–94. https://doi.org/10.1016/j.tox.2012.09.011

Mrema

E.J.

Rubino

F.M.

Brambilla

Moretto

, et al. “Persistent Organochlorinated Pesticides and Mechanisms of Their Toxicity”. Toxicology. 2013. 307: 74–88. https://doi.org/10.1016/j.tox.2012.11.015

Jayaraj

Megha

Sreedev

. “Organochlorine Pesticides, Their Toxic Effects on Living Organisms and Their Fate in the Environment”. Interdiscip Toxicol. 2016. 9(3–4): 90–100. https://doi.org/10.1515/intox-2016-0012

Taiwo

A.M.

. “A Review of Environmental and Health Effects of Organochlorine Pesticide Residues in Africa”. Chemosphere. 2019. 220: 1126–1140. https://doi.org/10.1016/j.chemosphere.2019.01.001

Lambropoulou

D.A.

Albanis

T.A.

. “Methods of Sample Preparation for Determination of Pesticide Residues in Food Matrices by Chromatography–Mass Spectrometry-Based Techniques: A Review”. Anal. Bioanal. Chem. 2007. 389(6): 1663–1683. https://doi.org/10.1007/s00216-007-1348-2

Samsidar

Siddiquee

Shaarani

S.M.

. “A Review of Extraction, Analytical, and Advanced Methods for Determination of Pesticides in Environment and Foodstuffs”. Trends Food Sci. Technol. 2018. 71: 188–201. https://doi.org/10.1016/j.tifs.2017.11.011

10.

Owen-Reece

Smith

Elwell

C.E.

Goldstone

J.C.

. “Near Infrared Spectroscopy”. Br. J. Anaesth. 1999. 82(3): 418–426. https://doi.org/10.1093/bja/82.3.418

11.

Long

D.A.

. Raman Spectroscopy. New York: McGraw-Hill, 1977. Pp. 310.

12.

Chen

Long

Song

Chen

Peng

, et al. “Rapid and Sensitive Detection of Pesticide Residues Using Dynamic Surface-Enhanced Raman Spectroscopy”. J. Raman Spectrosc. 2020. 51(4): 611–618. https://doi.org/10.1002/jrs.5823

13.

Lapcharoensuk

Fhaykamta

Anurak

Chadwut

Sitorus

. “Nondestructive Detection of Pesticide Residue (Chlorpyrifos) on Bok Choi (Brassica rapa subsp. Chinensis) Using a Portable NIR Spectrometer Coupled with a Machine Learning Approach”. Foods. 2023. 12(5): 955. https://doi.org/10.3390/foods12050955

14.

Ndung’u

C.N.

Kaniu

M.I.

Wanjohi

J.M.

. “Optimization of Diffuse Reflectance Spectroscopy Measurements for Direct and Rapid Screening of Pesticides: A Case Study of Spinach”. Spectrochim. Acta, Part A. 2022. 280: 121556. https://doi.org/10.1016/j.saa.2022.121556

15.

Ndung’u

C.N.

Kaniu

M.I.

Wanjohi

J.M.

Odongo

K.O.

, et al. “Feasibility for Rapid On-Site Screening of Pesticide Residues in Fresh Produce Using Machine Learning-Assisted Diffuse Reflectance Spectroscopy”. Food Humanity. 2024. 2: 100204. https://doi.org/10.1016/j.foohum.2023.100204

16.

Brereton

R.G.

Lloyd

G.R.

. “Partial Least Squares Discriminant Analysis: Taking the Magic Away”. J. Chemom. 2014. 28(4): 213–225. https://doi.org/10.1002/cem.2609

17.

Lee

L.C.

Liong

C.-Y.

Jemain

A.A.

. “Partial Least Squares-Discriminant Analysis (PLS-DA) for Classification of High-Dimensional (HD) Data: A Review of Contemporary Practice Strategies and Knowledge Gaps”. Analyst. 2018. 143(5): 3526–3539. https://doi.org/10.1039/c8an00599k

18.

de Juan

Jaumot

Tauler

. “Multivariate Curve Resolution (MCR): Solving the Mixture Analysis Problem”. Anal. Methods. 2014. 6(14): 4964–4976. https://doi.org/10.1039/C4AY00571F

19.

Ruckebusch

Blanchet

. “Multivariate Curve Resolution: A Review of Advanced and Tailored Applications and Challenges”. Anal. Chim. Acta. 2013. 765: 28–36. https://doi.org/10.1016/j.aca.2012.12.028

20.

Robert

I.I.O.

Birech

Kaniu

M.I.

. “Rapid Assessment of Molasses Adulterated Honey Using Laser Raman Spectroscopy and Principal Component Analysis”. Food Anal. Methods. 2023. 16: 1702–1710. https://doi.org/10.1007/s12161-023-02538-w

21.

Dhas

D.A.

Joe

I.H.

Roy

S.D.D.

Freeda

T.H.

. “DFT Computations and Spectroscopic Analysis of a Pesticide: Chlorothalonil”. Spectrochim. Acta, Part A. 2010. 77(1): 36–44. https://doi.org/10.1016/j.saa.2010.04.020

22.

Fleming

G.D.

Villagrán

Koch

. “IR, Raman, and SERS Spectral Analysis and DFT Calculations on the Herbicide O,S-Dimethyl Phosphoramidothioate, Metamidophos”. Spectrochim. Acta, Part A. 2013. 114: 120–128. https://doi.org/10.1016/j.saa.2013.05.012

23.

Huang

S.-G.

J.-P.

Liu

M.-H.

R.-M.

Wang

X.-B.

. “Density Functional Theory Calculation and Raman Spectroscopy Studies of Carbamate Pesticides”. Guangpuxue Yu Guangpu Fenxi [Spectrosc. Spectral Anal.]. 2017. 37(3): 766–771.

24.

Mao

Chen

, et al. “Vibrational Spectroscopy and Density Functional Theory Study of 3-[4,5-Dimethyl-2-thiazolyl]-2,5-diphenyl-2h-tetrazolium Bromide”. Spectrochim. Acta, Part A. 2015. 135: 1–6. https://doi.org/10.1016/j.saa.2014.06.084

25.

Wang

X.-B.

R.-M.

Liu

M.-H.

Zhang

L.-L.

, et al. “DFT and Raman Scattering Studies of Benzimidazole”. Guangpuxue Yu Guangpu Fenxi [Spectrosc. Spectral Anal.]. 2015. 35(6): 1562–1566.

26.

Lasch

Noda

. “Two-Dimensional Correlation Spectroscopy (2D-COS) for Analysis of Spatially Resolved Vibrational Spectra”. Appl. Spectrosc. 2019. 73(4): 359–379. https://doi.org/10.1177/0003702818819880

27.

Noda

. “Two-Dimensional Infrared Spectroscopy”. J. Am. Chem. Soc. 1989. 111(21): 8116–8118. https://doi.org/10.1021/ja00203a008

28.

Noda

. “Two-Dimensional Infrared (2D IR) Spectroscopy: Theory and Applications”. Appl. Spectrosc. 1990. 44(4): 550–561. https://doi.org/10.1366/0003702904087398

29.

Noda

. “Generalized Two-Dimensional Correlation Method Applicable to Infrared, Raman, and Other Types of Spectroscopy”. Appl. Spectrosc. 1993. 47(9): 1329–1336. https://doi.org/10.1366/0003702934067694

30.

Noda

. “Closer Examination of Two-Trace Two-Dimensional (2T2D) Correlation Spectroscopy”. J. Mol. Struct. 2020. 1213: 128194. https://doi.org/10.1016/j.molstruc.2020.128194

31.

Noda

Dowrey

A.E.

Marcott

, et al. “Generalized Two-Dimensional Correlation Spectroscopy”. Appl. Spectrosc. 2000. 54(7): 236A–248A. https://doi.org/10.1366/0003702001950454

32.

Park

Noda

Jung

Y.M.

. “Diverse Applications of Two-Dimensional Correlation Spectroscopy (2D-COS)”. Appl. Spectrosc. 2024. https://doi.org/10.1177/00037028241256397

33.

Geitner

Fritzsch

Popp

Bocklitz

T.W.

. “Corr2D: Implementation of Two-Dimensional Correlation Analysis in R”. J. Stat. Softw. 2019. 90(3): 1–33. https://doi.org/10.18637/jss.v090.i03

34.

Sheng

Xiao

, et al. “Simultaneous and Ultrasensitive Detection of Three Pesticides Using a Surface-Enhanced Raman Scattering-Based Lateral Flow Assay Test Strip”. Biosens. Bioelectron. 2021. 181: 113149. https://doi.org/10.1016/j.bios.2021.113149

35.

Yang

Xie

, et al. “Rapid Surface-Enhanced Raman Spectroscopy Detection of Chlorothalonil in Standard Solution and Orange Peels with Pretreatment of Ultraviolet Irradiation”. Bull. Environ. Contam. Toxicol. 2021. 107(2): 221–227. https://doi.org/10.1007/s00128-021-03258-9

36.

Abdi

Williams

L.J.

. “Principal Component Analysis”. WIREs Comput. Stat. 2010. 2(4): 433–459. https://doi.org/10.1002/wics.101

37.

Hearst

M.A.

Dumais

S.T.

Osuna

Platt

Scholkopf

. “Support Vector Machines”. IEEE Intell. Syst. Appl. 1998. 13(4): 18–28. https://doi.org/10.1109/5254.708428

38.

Hanson

B.A.

. “ChemoSpec: An R Package for the Chemometric Analysis of Spectroscopic Data (Package Version 4.4.97)”. http://cran.nexr.com/web/packages/ChemoSpec/vignettes/ChemoSpec.pdf [accessed Nov 12 2024].

39.

Dimitriadou

Hornik

Leisch

, et al. “Package e1071”. In: Miscellaneous Functions of the Department of Statistics, Probability Theory Group (Formerly E1071), TU Wien. Pp. 297–304. https://cran.r-project.org/web/packages/e1071/e1071.pdf [accessed Nov 12 2024].

40.

Grdadolnik

. “Infrared Difference Spectroscopy: Part I. Interpretation of the Difference Spectrum”. Vib. Spectrosc. 2003. 31(2): 279–288. https://doi.org/10.1016/S0924-2031(03)00018-3

41.

Zhang

Yang

. “Advanced Chemometrics Toward Robust Spectral Analysis for Fruit Quality Evaluation”. Trends Food Sci. Technol. 2024. 104612. https://doi.org/10.1016/j.tifs.2024.104612

42.

Zhang

Yang

, et al. “Near-Infrared Fluorescent Probe for Ultrasensitive Detection of Organophosphorus Pesticides and Visualization of Their Interaction with Butyrylcholinesterase in Living Cells”. Talanta. 2024. 279: 126587. https://doi.org/10.1016/j.talanta.2024.126587

43.

Sanchez

Pant

Irey

Mandadi

Kurouski

. “Detection and Identification of Canker and Blight on Orange Trees Using a Hand-Held Raman Spectrometer”. J. Raman Spectrosc. 2019. 50(12): 1875–1880. https://doi.org/10.1002/jrs.5741

44.

Farber

Shires

Ong

Byrne

Kuroski

. “Raman Spectroscopy as an Early Detection Tool for Rose Rosette Infection”. Planta. 2019. 250(4): 1247–1254. https://doi.org/10.1007/s00425-019-03216-0

45.

Dobrowolski

J.C.

. “Vibrational Spectroscopy as a Tool to Investigate Carotenoids”. In: Kaczor

Baranska

, editors. Carotenoids: Nutrition, Analysis, and Technology. Chichester, UK: Wiley–Blackwell, 2016. Chap. 6. Pp. 75–102.

46.

Marshall

C.P.

Olcott Marshall

. “The Potential of Raman Spectroscopy for the Analysis of Diagenetically Transformed Carotenoids”. Philos. Trans. R. Soc. Math. Phys. Eng. Sci. 2010. 368(1922): 3137–3144.

47.

Verma

S.P.

Hoelzl Wallach

D.F.

. “Carotenoids as Raman-Active Probes of Erythrocyte Membrane Structure”. Biochim. Biophys. Acta, Biomembr. 1975. 401(2): 168–176.

48.

Baranski

Baranska

Schulz

. “Changes in Carotenoid Content and Distribution in Living Plant Tissue Can Be Observed and Mapped In Situ Using NIR-FT-Raman Spectroscopy”. Planta. 2005. 222(3): 448–457. https://doi.org/10.1007/s00425-005-1566-9

49.

Schulz

Baranska

Baranski

. “Potential of NIR-FT-Raman Spectroscopy in Natural Carotenoid Analysis”. Biopolymers. 2005. 77(4): 212–221.

50.

Agarwal

U.P.

. “1064 nm FT-Raman Spectroscopy for Investigations of Plant Cell Walls and Other Biomass Materials”. Front. Plant Sci. 2014. 5: 490. https://doi.org/10.3389/fpls.2014.00490

51.

Mary

Y.S.

Panicker

C.Y.

Varghese

H.T.

. “Vibrational Spectroscopic Investigations of 4-Nitropyrocatechol”. Orient. J. Chem. 2012. 28(2): 937–941.

52.

M.M.

Schulze

H.G.

Jetter

Blades

M.W.

Turner

R.F.B.

. “Raman Microspectroscopic Analysis of Triterpenoids Found in Plant Cuticles”. Appl. Spectrosc. 2007. 61(1): 32–37. https://doi.org/10.1366/000370207779701352

53.

De Oliveira

V.E.

Castro

H.V.

Edwards

H.G.M.

de Oliveira

L.F.C.

. “Carotenes and Carotenoids in Natural Biological Samples: A Raman Spectroscopic Analysis”. J. Raman Spectrosc. 2010. 41(6): 642–650. https://doi.org/10.1002/jrs.2493

54.

Agarwal

U.P.

. “Raman Imaging to Investigate Ultrastructure and Composition of Plant Cell Walls: Distribution of Lignin and Cellulose in Black Spruce Wood (Picea mariana)”. Planta. 2006. 224(5): 1141–1153. https://doi.org/10.1007/s00425-006-0295-z

55.

Kang

Wang

Zou

. “High Pressure Structural Investigation of Benzoic Acid: Raman Spectroscopy and X-ray Diffraction”. J. Phys. Chem. C. 2016. 120(27): 14758–14766. https://doi.org/10.1021/acs.jpcc.6b05001

Enhanced Detection of Pesticide Residues Using Two-Dimensional Raman Correlation Spectroscopy and Machine Learning

Abstract

Keywords

Introduction

Experimental

Materials and Methods

Sample Preparation, Data Acquisition, and Pre-Processing

Two-Dimensional Correlation Spectroscopy (2D-COS) Analysis and Machine Learning Model Development

Results and Discussion

Raman Spectral Analysis of Vegetable Samples Highlighting Bio-Molecular and Chlorothalonil-Specific Vibrational Peaks

Identifying Chlorothalonil Fingerprints and Spectral Variations in Raman Spectra

Difference Spectroscopy Results

Enhanced Resolution of Chlorothalonil Fingerprints Using 2D-COS

Principal Component Analysis (PCA)-SVM Results for Classifying Spiked and Control Samples

Principal Component Analysis (PCA)-SVM Results for the Full Spectral Range

Principal Component Analysis (PCA)-SVM Results Using Identified Fingerprints

Conclusion

Footnotes

Acknowledgments

Author Contributions

Data Availability Statement

Declaration of Conflicting Interests

Declaration of Generative Artificial Intelligence (AI) and AI-Assisted Technologies in the Writing Process

Funding

ORCID iDs

References