Carbon dots from Manila tamarind for heavy metal ion removal and sensing through automatic classification
Simei Darinel Torres Landa, Luis Felipe Ávalos Ruiz, José Francisco Gómez Aguilar, Vivechana Agarwal
TL;DR
This paper introduces carbon dots made from Manila tamarind that can detect and remove heavy metal ions, using optical signals and machine learning for environmental cleanup.
Contribution
The novelty lies in using Manila tamarind-derived carbon dots for dual sensing and removal of heavy metals with an 'on-off-on' strategy and machine learning classification.
Findings
Carbon dots effectively detect and remove Fe3+, Pb2+, and Sn2+ through electron transfer and electrostatic interactions.
Machine learning algorithms can identify 17 metal ions using spectral and statistical features of the carbon dots.
XPS analysis confirms the reduction of Fe3+ to Fe+2 and Pb2+ to Pb0 states by the carbon dots.
Abstract
Environmental remediation research has been focused on the detection and removal of heavy metal ions. In this work, Manila tamarind-derived simple and sustainable carbon dots (CDs) have been proposed for the optical detection/removal of heavy metal ions. Based on the distinct optical response of CDs toward Fe3+ and Zn2+/paraquat, an “on-off-on” strategy could be implemented for dual sensing of Fe3+/Zn+2 and Fe3+/paraquat systems. Higher concentrations of CDs revealed the facile removal of Fe3+ and Pb2+ through electron transfer mechanism and of Sn2+ possibly via electrostatic interaction. X-ray photoelectron spectroscopy (XPS) reveals the CD-induced reduction of Fe3+ and Pb2+ ions to Fe+2 and Pb0 states, respectively. The spectral and statistical features were analyzed for possible identification of 17 different metal ions through machine learning approaches. The demonstrated…
Genes, proteins, chemicals, diseases, species, mutations and cell lines named across the full text — each resolved to its canonical identifier and authoritative record.
Click any figure to enlarge with its caption.
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7
Figure 8
Figure 9
Figure 10Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsCarbon and Quantum Dots Applications · Electrochemical sensors and biosensors · Melamine detection and toxicity
Introduction
The contamination of water by heavy metal ions can lead to environmental issues and adverse impacts on human health. Wastewater containing traces of Pb^2+^, originating from industrial processes, poses significant risks to human health. The contact with water contaminated by Pb^2+^ can result in kidney damage, neurological disorders, reproductive health problems, cardiovascular diseases, anemia, or even fatality.1^,^2 Sn^2+^ ions are naturally found in the human body, however; a deficiency of these ions is linked to hearing impairment and inadequate bodily growth, whereas an excess can lead to impairments in the digestive, respiratory, and nervous systems.3^,^4 Iron and zinc are the most abundant and essential heavy metals in the human body; however, their presence at elevated levels in drinking water results in a large number of health problems. Perturbations of Zn^2+^ levels in humans can lead to neurological disorders such as, episodes of schizophrenia, Alzheimer's disease, and Parkinson's disease. The excess of Zn^2+^ in soil can induce a phytotoxic effect, which disrupt the crop growth.5^,^6 The excess or deficiency of Fe^3+^ can result in malfunctioning of organisms, causing Parkinson disease, Alzheimer disease, or anemia.7 Aside from heavy metal ions, environmental issues may also center on the detrimental impacts of various pesticides. One of the most frequently utilized pesticides is the toxic paraquat (N,N′-dimethyl-4,4′-bipyridinium dichloride, with a molecular formula of C_12_H_14_Cl_2_N_2_ and a molecular weight of 257.2 g/mol).8 Ingestion of paraquat can cause fibrosis in the lungs, respiratory damage, and, due to the absence of drug treatment for pesticide intoxication, has resulted in high mortality in in vivo studies.9
The detection and elimination of such water pollutants are crucial for environmental remediation and protection of human health. Apart from various conventional techniques, several innovative materials have been investigated for removing heavy metal ions from polluted water. For example, elimination of Pb^2+^ through biosorbents derived from agro-waste and bacterial biomass,10 ball-milled magnetic biochar derived from bone,11 or a hybrid cation exchange material, PANI-Ti(IV) phosphosulphosalicylate,12 has been reported. Additionally, for the removal of Fe^3+^, adsorbents based on thiol-functionalized zirconium-based metal-organic frameworks,13 captors/adsorbents of hierarchical, microscopic structures of TiO_2_ and TiOF_2_,14 or nanofibrous composites of chitosan/(polyvinyl alcohol)/zeolite15 have been proposed. Sn^2+^ have been removed from water by using nano-crystalline calcium hydroxyapatite16 and ZnFe_2_O_4_-carbon nanotube adsorbent material,17 and poly(4,4′-biphenol oxalate) oligomer has been also explored as a metal-ion uptaker.10
Furthermore, the emerging class of carbon dot (CD)-based nanomaterials has also been explored due to their simple synthesis, outstanding luminescence, water solubility, and biocompatible nature.18^,^19^,^20 The distinctive properties of CDs have been widely used in bioimaging,21 drug delivery,22^,^23 anti-bacterial applications,24 optical sensing,25^,^26 among other applications. Even though some green CD-based independent optical sensors have been reported for Fe^3+^,27^,^28^,^29^,^30 Sn^2+^,31^,^32^,^33 Zn^2+^,5^,^34^,^35 or paraquat detection,9^,^36 the removal of heavy metal ions remains largely an unaddressed issue.
The current research introduces green and sustainable Manila tamarind (a vastly invasive plant, also known as Pithecellobium dulce)-derived CDs to detect various heavy metal ions selectively. This method features distinct identification capabilities for Fe^3+^, Sn^2+^ (exhibiting a simple turn-off/turn-off with redshift), and Zn^2+^ (showing turn-on with blueshift). Here, the possible implementation of “on-off-on” bimodal detection systems implemented between Fe^3+^-Zn^2+^ and between Fe^3+^-paraquat as luminescence quencher-activator was analyzed. Additionally, at higher concentrations of CDs, the precipitation of heavy metal ions opens the opportunity for the removal of highly toxic heavy metals (Pb^2+^, Fe^3+^, and Sn^2+^) from water, which is accompanied by the reduction of Fe^+3^ and Pb^2+^, suggesting that this method is a promising technique with potential environmental applications. In summary, a multifunctional sensor has been developed for the selective detection of multiple heavy metal ions, with the additional ability to remove and reduce certain toxic metal ions from water. Finally, the feasibility of the CD-based optical sensor was further assessed by identifying specific features of signals to differentiate between the metal ions by using machine learning models.
Results
Optical and physicochemical characterization
Figure 1A) displays the ultraviolet-visible absorption, photoluminescence excitation (PLE), and emission spectra of CDs_400_. In the UV-vis spectra, two distinct absorption peaks at 260 nm and 325 nm were observed; these peaks were ascribed to the π→π^∗^ transition of conjugated C=C and to the n→π^∗^ transition associated with C=O, respectively. The broad emission band with a PL_max_ located at 406 nm confirms the blue emission (inset) from CDs_400_. The average particle size of the as-prepared CD_400_, characterized by TEM (Figure 1B), was estimated to be ∼5 ± 1.3 nm. The corresponding FTIR spectrum (Figure 1C) reveals the different bands, i.e., -OH at ∼3,660 cm^−1^,37 C–H at 2,901 cm^−1^,38 C=O located at 1,650 cm^−1^,39 C–O–H/C=C signal at 1,415 cm^−1^,40 and C-O and C-H at 1,066 and 880 cm^−1^, respectively.41^,^42Figure 1. Basic characterization of carbon dots(A) Absorbance, PL, and PLE.(B) TEM image of the as-prepared CD_400_ (inset: size distribution from 20 particles). Scale bars, 20 nm.(C) FTIR spectrum and XPS results.(D) Survey (CDs_400_ compared with CDs_200_).(E) High-resolution spectra of C 1s, O 1s, and N 1s for CDs obtained at 400°C.
The surface elements and chemical states of CDs were analyzed by using X-ray photoelectron spectroscopy (XPS). The three characteristic peaks of C 1s (284.6 eV), O 1s (530.9 eV), and N 1s (399.2 eV) were observed for CDs synthesized at 400°C (Figure 1D). Besides, signals corresponding to K (377 and 392 eV) and Ca (439.1, 351, and 347.1 eV) are shown in the XPS survey. These signals could be related to Manila tamarind, which contains vitamins, calcium, phosphorous, and iron.43 Calcium composition as calcite (CaCO_3_) has been reported to enhance the biosorption capacity of heavy metals.44^,^45 The high-resolution spectrum of C 1s was deconvoluted into three peaks—284.3, 284.9, and 286.9 eV (C–O, C–C/C=C, and O–C=O, respectively). The O 1s spectrum displayed two bands at 530.9 and 531.7 eV (C–O/C=O and C–OH/C–O–C, respectively), while the N 1s spectrum showed a main peak at 399.2 eV and a slight signal at 403 eV, which could be attributed to some protonated nitrogen atoms (Figure 1E).46 Similar characterizations of the as-prepared CDs_200_ are presented in Figures S1A–S1E. A comparison of the fluorescence stability of CDs under different pH values and NaCl (0.5–4 M) is presented in Figures S1F and S1G.
Specificity toward metal ions
Although specificity tests conducted on both CDs_200_ and CDs_400_ (Figure 2) demonstrate a response to Zn^2+^, Fe^3+^, and Sn^2+^, the nature of the response distinguishes each metal uniquely. For CD_200_ (Figures 2A and 2B), an increase in the photoluminescence signal (PL), accompanied by a blue shift (approximately 10 nm) in the presence of Zn^2+^ is observed. On the other hand, a decrease in luminescence intensity accompanied with a redshift (approximately 13 nm) is observed for Fe^3+^. Similar selectivity observations made for CDs_400_ (Figures 2C and 2D) toward different monovalent, bivalent, and trivalent metal ions reveal strong selectivity toward Sn^2+^ as the decrease in luminescence intensity is accompanied by a redshift of ∼19 nm. Although both types of CDs demonstrate an increase in PL signal (with blueshift) in the presence of Zn^2+^, along with a decrease in the corresponding signal with Fe^3+^ (red shifted by ∼20 nm), the displacement with Zn^2+^ is relatively less in CDs_400_ than in CDs_200_. The PL spectra in Figures S2A–S2D reveal the influence of anions in the selectivity of CDs. In the case of CuCl_2_ or Cu(NO_3_)2, CdCl_2_ or Cd(NO_3)2, and CoCl_2 or CoSO_4_, the maximal PL signal remains in the same wavelength with a slight change in intensity. Figure S7 shows the normalized confusion matrix illustrating the classification performance of the proposed model, where classes are grouped according to the counter anion while disregarding the specific metal cation. Notably, chlorides achieve near-perfect identification accuracy. In PL, an exception is observed for iron, where FeCl_3_ exhibits a complete turn-off effect relative to Fe_2_(SO)4.Figure 2. Specificity of CDs toward different metal ions(A and C) Specificities of CDs_200_ and CDs_400_, respectively.(B and D) Corresponding relative PL_max_ signals for each metal ion with respect to the signal of CDs in deionized water (F/F_0_). Error bars represent SD of tripled measurements.
The quenching behavior of Fe^3+^, on one hand, is analyzed by testing the sensitivity of CDs_200_ toward Fe^3+^ in the range of 1 nM–10 mM (Figure 3A). High concentrations of Fe^3+^ ion display a redshift. The linear equation fitted as with a linear correlation (R^2^ = 0.97) in the range of 100 nM to 100 μM (Figure 3B) and an LOD of 11 μM. On the other hand, sensitivity measurements performed using CDs_200_ (Figure 3C), at different concentrations of Zn^+2^, reveal an opposite tendency with respect to Fe^3+^, i.e., an increasing PL signal with an increase in Zn^2+^ concentration (turn-on mechanism), also accompanied by a shift of ∼12 nm toward higher energy emission values. The resulting linear relationship is expressed as F/F0 = 0.9921–0.0102[Zn^2+^], with an R^2^ = 0.99 (Figure 3D) at lower concentrations of M^+^ (metal ions). The sensitivity of CD_400_ to Sn^2+^ ions (Figure 3E) also revealed a dual signal, i.e., the decrease in fluorescence intensity with a chromatic red shift of around 11 nm (from 406 to 420 nm). The linear adjustment was settled as F⁄F0 = 0.8098 + 0.0052[Sn^2+^], with an R^2^ = 0.97 (Figure 3F).Figure 3. Sensitivity measurements for CD_200_ and CD_400_(A) CD_200_, PL signal vs. wavelength as a function of Fe^3+^ concentration.(B) Linear adjustment of relative PL_max_ intensity with Fe^3+^.(C) CDs_200_, PL signal vs. wavelength as a function of Zn^2+^ concentration.(D) Linear fit to Zn^2+^ detection.(E and F) Sensitivity of CDs_400_ to Sn^2+^ ions (E) and its linear adjustment (F). Error bars represent SD of tripled measurements.
Taking into consideration the opposite nature of the optical response of CDs in the presence of Fe^3+^ and Zn^2+^, an “on-off-on” system could be designed to detect the two metal ions simultaneously. Figures 4A and 4B show the quenching with CDs/Fe^3+^ [100 μM] system (turn-off) that was turned-on to the initial PL intensity (of CDs) with the addition of Zn^2+^(1.8 mM in the present case). The LOD was determined to be 14 μM, and the Stern-Volmer equation is represented by F/F_0_ = 1.0871–0.0030[Zn^2+^], with an R^2^ = 0.99 (Figure 4C). Similarly, Figures 4D and 4E demonstrate the recovered PL signal of the CDs/Fe^3+^ [100 μM] system in the presence of the pesticide paraquat (from 8 to 100 μM).Figure 4. On-off-on system studies(A) PL recovery of the CDs/Fe^3+^ system with the addition of Zn^2+^.(B) Bar graph for PL recovery with Zn^2+^ addition, showing changes in λmax intensity.(C) Linear adjustment.(D) PL recovery of CD/Fe^3+^ system with the addition of paraquat.(E) Bar graph for PL recovery with paraquat addition, showing changes in λmax intensity. Error bars represent SD of tripled measurements.
In addition to the optical detection at low concentrations of CDs, an increase in CD concentration leads to the possible complexation of Fe^3+^, Pb^2+^, and Sn^2+^ with the CDs. Figure 5 clearly shows the sediments formed (CDs/metal ion) for CD_200_ and CD_400_. Removal percentages for CDs_200_ in deionized water and in lake water are shown in Figures S3A–S3F. The removal efficiencies (%R) for Fe^3+^, Pb^2+^, and Sn^2+^ in deionized water were 94%, 96%, and 96%, respectively. In lake water, these efficiencies decreased to 75%, 83%, and 83%, respectively, at a metal ion concentration of 830 μM. When the metal ion concentration was varied from 830 μM to 5.8 mM, only slight changes in %R were observed. The reusability of the CDs was explored; however, the results indicate that after their initial use for Fe^3+^ removal, the supernatant solution containing CDs no longer exhibited significant effectiveness in further iron removal (Figure S4).Figure 5. Photographic images under daylight and UV light (365 nm)(A) Increasing concentration of carbon dots for CD_200_ and CD_400_.(B) Naked eye observations of metal ions as their concentration increases from 160 μM to 5.8 mM (from left to right) for both types of CDs.
The distinct morphology of precipitates resulting from the interaction between CDs_200_ and metal ions was investigated using Field Emission Scanning Electron microscope (FE-SEM). The precipitate formed with CDs_200_/Fe^3+^ reveals the configuration of nanorods (Figure 6A). While the CDs_200_/Pb^2+^ sediments contain clusters with an average dimension of ∼200 nm (Figure 6B), the sediments of CDs_200_/Sn^2+^ display agglomerations of nanoparticles with an average dimension of ∼100 nm (Figure 6C). To understand the elemental conformation and the chemical state, the precipitates were removed from the solutions and investigated using XPS. The survey spectrum verified the presence of Fe^3+^, Pb^2+^ and Sn^2+^ in the carbonaceous micromorphologies (Figures 6D–6F). The deconvoluted peaks, from the iron-CDs_200_ XPS analysis, at 709 and 723.8 eV are associated with Fe^2+^, while the peaks of 712.9 and 734.2 eV are related to Fe^3+^, as shown in Figure 6G.47 The C1s spectrum (Figure S5A) displays five peaks—283.3, 284.7, 286, 291, and 294.1 eV—ascribed to C–H, C–C/C=C, C=O, O–C=O/C–N, and π-π^∗^, respectively.48^,^49 High-resolution spectrum of O1s (Figure S5B) was resolved into three signals at 529.5, 530.5, and 533 eV from O–C, C=O, and HO–C/C–O–C, respectively, while N 1s spectra (Figure S5C) were deconvoluted into two peaks at 398.5 and 400.2 eV, which could be related to NH_2_ and N–C, respectively.50 High-resolution spectrum of Pb^2+^ was deconvoluted into four signals at 138.9 and 143.5 eV, assigned to Pb^2+^, and at 138.2 and 141.7 eV from Pb^0^ (Figure 6H).51 In addition, two peaks at 284.7 from C–C/C=C bonds and at 287.9 eV from the C=O were observed in the C1s spectrum (Figure S5D). The O1s spectrum reveals the C=O signal at 530.3 eV and C–O at 530.6 eV (Figure S5E). In this case, the nitrogen signal could overlap with the Pb4d peak at around 400 eV. The peaks corresponding to Sn 3d_5/2_ and 3d_3/2_ were displayed at 486.9 and 495 eV52 in the high-resolution spectra (Figure 6I). The O1s spectrum showed two resolvable peaks at 531.8 and 533.4 eV, attributed to C=O and C–O bands (Figure S5F), respectively.53 The C1s spectrum was resolved into five peaks assigned to C–H (285.4 eV), C–C/C=C (286.9 eV), C=O (288.1 eV), O–C=O/C–N (292.9 eV) (Figure S5G), and π–π^∗^ (295.8 eV). The enlarged signal corresponding to N 1s was revealed at ∼400 eV (Figure S5H). Energy-dispersive X-ray spectroscopy (EDS) analysis verifies the elemental composition of the precipitates formed by each tested metal ion by using CDs_200_ and CDs_400_ (Figure S6).Figure 6FE-SEM images of heavy metals precipitated with CDs_200_ and corresponding XPS survey results(A) FE-SEM image for CDs_200_/Fe^3+^. Scale bars, 500 nm.(B) FE-SEM image for CDs_200_/Pb^2+^. Scale bars, 500 nm.(C) FE-SEM image for CDs_200_/Sn^2+^. Scale bars, 1 μm.(D–F) XPS survey results for (D) CDs/Fe^3+^, (E) CDs/Pb^2+^, and (F) CDs/Sn^2+^.(G–I) High-resolution spectra of (G) Fe 2p, (H) Pb 4f, and (I) Sn 3d.
Application in detection/reduction/removal
Through surface complexation, electrostatic interactions, or electron transfer mechanisms, CDs engage with metal ions, altering their optical, electrical, and catalytic characteristics. This tunability facilitates their use in metal detection, reduction, and removal, with promising applications in water treatment and quality monitoring.54 Table 1 presents a comparative analysis of carbon-based nanomaterials employed for metal ion detection, reduction, and removal, highlighting the superior performance of the CDs synthesized in this study.Table 1. Comparison of different carbon-based system for detection, reduction, or removal of some metal ionsMaterialSystem arrangementMetal ion detectedLOD (μM)Metal ions removed or reduced% removalReductionReferenceChitosan-montmorillonite systemcomplex––Cu^2+^Ni^2+^Co^2+^∼100∼83∼76–Assaad et al.55CDs from chitosansimpleAg^+^0.13Ag^+^–Ag^0^Shen et al.56CDs from onion and grapesimpleFe^3+^0.1Fe^3+^–Fe^2+^Shariati-Rad et al.57Carboxymethyl chitosan-MACA (mercapto alkyl carboxylic acid) copolymercomplex––Ni^2+^∼94–Sun et al.58CDs/Al_2_O_3_ nanofibers2020complex––Pb^2+^∼99–Fouda-Mbanga et al.59CDs from carrotsimpleCu^2+^0.187Cu^2+^Pb^2+^Ag^+^Al^3+^Cr^3+^Fe^3+^Hg^+2^––Liu et al.60CDs from avocado seedssimpleCu^2+^Cr^6+^0.00247x10^−5^Cu^2+^∼98–Mejía Ávila et al.61CD-doped hydrogel particlescomplex––Pb^2+^Hg^2+^Cd^2+^Cr^3+^>70∼99>70>70–Perumal et al.62Co-polymer of maleic acid-acrylic acid (PMA)complex––Sn^2+^∼99–Le et al.63CDs from cambuci juicesimpleZn^2+^5.4–––Da Silva Júnior et al.64CDs from Manila tamarind (this work*)*simpleFe^3+^Sn^2+^Zn^2+^paraquat11201.9Fe^3+^Sn^2+^Pb^2+^∼94∼96∼96Fe^2+^Pb^0^this work
Possible sensing and removal mechanism
CDs exhibit a bivalent redox nature, which enables them to function as both electron donors and electron acceptors. Their surface, enriched with diverse functional groups such as carboxyl, hydroxyl, and amine, facilitates the binding of electron donors or acceptors through covalent or non-covalent interactions.65^,^66 In addition, owing to the electron-donating properties of CDs, the chemical electron transfer pathway may facilitate the reduction of Fe^3+^ and Pb^2+^, like Fe^3+^+e^−^→,Fe^2+^ and Pb^2+^+2e^−^→Pb^0^ (as confirmed by XPS results), and the oxidation of the CDs.67^,^68 The high-resolution spectrum of O1s of CDs_200_-Fe precipitates shows a new band corresponding to HO–C/C–O–C bonds, as well as an increased in the intensity of the C=O band, and the C1s spectrum also displays considerable shifts in comparison with the pristine CDs, confirming their surface chemistry modification due to the oxidation of CDs owing to their interaction with Fe^3^ ions (Figures S1 and S4). On the other hand, the observed luminescence quenching has been linked to electron transfer from the excited state of CDs to the heavy metal ions,69 as represented in Figure 7. For example, Shariati-Rad et al. suggested the reducing property of grape-/onion*-*derived CDs, which converted Fe^3+^ to Fe^2+^, due to the hydroxyl groups (–OH) on the surface of the synthesized CDs. The –OH could promote the formation of carbonyl groups.57Figure 7. Schematic of Fe^3+^ reduction and removal mechanism
The recovery/removal of lead from the Pb^2+^-spiked water is ascribed to a possible electron transfer mechanism, as evidenced by XPS, i.e., CDs could possibly facilitate the reduction of Pb^2+^ to Pb^0^ as well as promote the precipitation process. Moreover, the O1s high-resolution XPS spectrum from CDs was deconvoluted into one band from C–O/C=O (530.9 eV) and the other related to C–OH/C–O–C (531.7 eV) bonds, while the O1s spectra corresponding to the CDs/Pb^2+^ nanoparticles show only signals attributed to C–O/C=O (530.7/530.3 eV). This alteration in the oxygen-containing functional groups suggests the oxidation action of CDs. In addition, the C1s high-resolution XPS spectra of precipitates show two bands in comparison with the four kinds of bonds in the spectrum of pristine CDs (Figures S1 and S4). In conformity with the work by Liu et al., the C–OH band could be linked to a hydroxyl interaction in Pb^2+^ reduction.70 However, Gettrick et al. found that the Pb^2+^ reduction could be due to the XPS energy (in case of perovskite films) and concluded that radiolysis could form the Pb^0^ clusters.49 To address this issue, a control and CDs from Manila tamarind were simultaneously tested in the Pb^2+^ precipitation experiments. CDs derived from Manila tamarind demonstrated metal reduction activity.
The negative charge of CDs from Manila tamarind (∼-20 eV, due to nitrogen and oxygen functional groups) is oppositive to the charge of metal ions; hence, the electrostatic interaction is one of the most common and fundamental aspects associated with precipitation and could be the most probable mechanism associated with the removal and PL quenching of Sn^2+^.
Machine learning for the identification of M+
A machine learning approach was employed to automatically identify heavy metal ions based on specificity tests conducted for CDs_200_ and CDs_400_. To reduce noise, a moving average filter with a window size of five samples was applied. The signal amplitudes were normalized to the [0,1] range, while the wavelength axis was truncated to [380,680] nm and subsequently normalized.
A total of 17 statistical and spectral features were analyzed. The statistical features included the maximum and minimum amplitude values and their corresponding wavelengths, along with the mean, standard deviation, skewness, and kurtosis of the signal. Additionally, energy-based features were computed in both the wavelength and frequency domains. In the wavelength domain, energy was calculated over predefined spectral bands, with boundaries at 380, 450, 485, 500, 565, 590, 625, and 750 nm, corresponding to distinct color regions in the visible spectrum. In the frequency domain, the Fourier transform was computed, and the spectrum was divided into five equal bands after discarding the second half of the symmetric representation. Energy was then extracted from each of these frequency bands. Feature selection was performed using neighborhood component analysis (NCA) to identify the most relevant predictors for classification.71 The analysis determined that energy in the first frequency band, skewness, and kurtosis were the most significant features. This selection process effectively reduced dimensionality while preserving key discriminatory information.
To augment the original dataset, synthetic samples were generated based on the cluster structure identified within the feature space. Each of the 17 classes exhibited two distinct clusters, which were treated separately during the augmentation process. For each cluster, the mean and standard deviation were calculated, and new data points were sampled from a Gaussian distribution centered at the cluster mean, using the corresponding standard deviation as the scale parameter. Figure 8 displays the distribution of different metal ions and water (reference/control) based on the selected features.Figure 8. Predictors selected for classification based on NCA for each of the heavy metal ionsThe selected heavy metal ions are (1) Cu^2+^, Cd^2+^, Pb^2+^, (1) Co^2+^, Zn^2+^, Sn^2+^, Hg^2+^, Fe^2+^, (2) Cu^2+^, Ni^2+^ (2) Co^2+^ (3) Cu^2+^, Cd^2+^, Ag^+^, Mg^2+^, Ba^2+^, and Fe^2+^.
Machine learning models were selected to demonstrate different methodologies.72 Support vector machines (SVMs) were selected due to their effectiveness in high-dimensional feature spaces and their capability to accommodate limited datasets, particularly through the utilization of kernel methods. Decision trees (DTs) were incorporated for their inherent interpretability. K-means clustering was employed to illustrate its potential for partitioning data into distinct clusters based on similarity measures. Each model was trained using the features obtained from the NCA, namely energy in the lowest frequency band, kurtosis, and skewness, and subsequently compared according to their accuracy, calculated as . Figure 9 shows the confusion matrix for the best-performing model, which achieved an accuracy of 95.6%. Misclassifications are notable in classes corresponding to different clusters within the same ion type. For instance, substantial confusion is observed between (2) Cu^2+^ and (3) Cu^2+^, as well as between the clusters associated with Cd^2+^. In the context of environmental monitoring,73^,^74 distinguishing between chemically similar subtypes may be less critical than ensuring accurate detection of contaminants. The confusion matrix indicates that most classes were successfully classified with a relatively low number of misclassifications. Further work involving a real-world dataset could provide additional insight into the generalizability of the model.Figure 9. Confusion matrix for SVM model with a fine Gaussian kernel function Acc = 95.6%
Discussion
Luminescent CDs were synthesized using leaves of Manila tamarind as the carbon source, employing a sustainable and facile strategy involving carbonization at two distinct temperatures (200°C and 400°C). The synthesis temperature was found to influence not only the particle size but also the CDs’ selectivity toward metal ions. Among the 17 metal ions tested, Fe^+3^, Sn^+2^, and Zn^+2^ were distinctly identified through the characteristic optical response of the CDs in their presence. Fe^3+^ ions were found to completely quench the luminescence of CDs_200_ (turn-off), whereas the presence of Zn^2+^ ions resulted in an enhancement of the emission signal characterized by a blue shift as a function of Zn^+2^ concentration. Additionally, the quenching of CDs_400_ luminescence in the presence of Sn^2+^ ions was characterized by a simultaneous redshift as a function of concentration. An “on-off-on” system was designed for the simultaneous detection of Fe^+3^ and Zn^+2^, capitalizing on their contrasting effects on the optical signal. A similar system was developed for the detection of the highly toxic and poisonous pesticide, paraquat, within a specific concentration range. Furthermore, higher concentrations of the CDs could be utilized for the precipitation and potential removal of Fe^3+^, Pb^2+^, and Sn^2+^ from water. Depending on the metal ion, the formation of different micromorphologies of the sediments was analyzed. The XPS analysis confirmed the simultaneous reduction of Fe^3+^ to Fe^+2^ and Pb^2+^ to Pb^0^ in the sediments, indicating the CDs’ potential role as a green reducing agent via electron transfer. With respect to the other systems reported in the literature, the present study reveals the multifaceted applications of these eco-friendly, cost-effective, and sustainable CDs as a green reducing agent, precipitation mediator, and optical sensor for metal ions in water, which suggest their promising applications in environmental detection and remediation. Future work will focus on extending the proposed methodology to complete interference studies in lab samples and real-world samples. Furthermore, dataset enrichment will improve the robustness of the machine learning model in real-world scenarios, considering interferents and concentration estimation in complex matrices.
NCA applied to the extracted features suggests that low-frequency signal components and distribution characteristics (kurtosis and skewness) are effective for differentiating heavy metals based on specificity. The SVM with a fine Gaussian kernel function and trained on a limited dataset achieved an accuracy of 95.6%, demonstrating the feasibility of applying this methodology in a scaled-up scenario.
Limitations of the study
A study of real, complex samples (including noise from polluted water) to analyze interference with the reported optical signals can further expand the scope of the work. An additional analysis of the interaction mechanisms using density functional theory (DFT) calculations can enhance understanding of the system and enable the design of similar systems. The implementation of standardized guidelines for water sample collection can strengthen the study. On the other hand, the machine learning approach used in this study relies on a synthetically augmented dataset generated via Gaussian sampling from the observed cluster statistics. This approach, while effective for expanding the dataset, may not fully represent the complexity and variability present in real-world conditions. Furthermore, certain misclassification patterns appear to stem from intrinsic similarities in the feature representations of ions belonging to the same metal species. Consequently, additional validation using real-world samples is required to assess the model’s generalizability more comprehensively.
Resource availability
Lead contact
- •Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Vivechana Agarwal ([email protected]).
Materials availability
- •This study did not generate new unique reagents, materials, or biological resources. Optical sensor materials and carbon dots were synthesized using standard laboratory procedures as described in STAR Methods.
Data and code availability
- •All data generated or analyzed during this study are included in the article and its supplemental information files. No large-scale datasets or external repositories were required for this non-biological study.
- •This study did not generate or use any custom code or software; therefore, no code is available.
- •No accession codes apply. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.
Acknowledgments
V.A. acknowledges the support from SNII-Secretaría de Ciencias, Humanidades, Tecnología e Innovación (SECIHTI). L.F. Avalos-Ruiz acknowledges the financial support provided by the SECIHTI through the postdoctoral fellowship program 2025 (No. 888206). The research of J.F. Gómez Aguilar was supported by SNII-SECIHTI.
Author contributions
Conceptualization, S.D.T.L. and V.A.; methodology, S.D.T.L. and V.A.; data curation, S.D.T.L.; writing – original draft, data analysis, data interpretation, and discussion, S.D.T.L. and L.F.Á.-R.; supervision and writing – review & editing, J.F.G.A. and V.A.; resources, V.A.; funding acquisition, V.A.
Declaration of interests
The authors declare no competing interest.
STAR★Methods
Key resources table
REAGENT or RESOURCESOURCEIDENTIFIERChemicals, peptides, and recombinant proteinsCuSO_4_Sigma AldrichCAT# 12852CoSO_4_Sigma AldrichCAT# 940151NiSO_4_Sigma AldrichCAT# 6568895FeSO_4_Sigma AldrichCAT#935689Cd(NO_3_)2_Sigma AldrichCAT# 642045Cu(NO_3)2_Sigma AldrichCAT#61194Mg(NO_3)2_Sigma AldrichCAT# 237175AgNO_3_Sigma AldrichCAT#209139Ba(NO_3)_2_Sigma AldrichCAT#217581ZnCl_2_Sigma AldrichCAT#208086FeCl_3_Sigma AldrichCAT#908908FeCl_2_Sigma AldrichCAT#372870SnCl_2_Sigma AldrichCAT#208256PbCl_2_Sigma AldrichCAT#268690CuCl_2_Sigma AldrichCAT#222011CoCl_2_Sigma AldrichCAT#232696HgCl_2_Sigma AldrichCAT#215465Software and algorithmsOrigin 2021bOriginlabhttps://www.originlab.com/Matlab 2022bMathWorkshttps://www.mathworks.com/BiorenderBiorenderhttps://www.biorender.com/OtherUV–Vis (Lambda 950)Perkin Elmerhttps://www.perkinelmer.com/Varian 660-IRAgilenthttps://www.agilent.com/TEM-JEM-2010 imagesJEOLhttps://www.jeol.com/X-ray photoelectron spectrometer, scalab 250XiThermo Fisherhttps://www.thermofisher.com/FE-SEM (Hitachi SU5000)Hitachihttps://www.hitachi-hightech.com/Spectrofluorometer (Cary Eclipse)Varianhttps://www.agilent.com/
Method details
Heavy metals of copper sulfate -CuSO_4_, cobalt sulphate -CoSO_4_ (99%), nickel sulfate -NiSO_4_ (98%), ferrous sulfate (II) -FeSO_4_ (99%), cadmium nitrate -Cd(NO_3_)2 (99%), cooper nitrate Cu(NO_3_)2, magnesium nitrate-Mg(NO_3_)2 (99.5%), silver nitrate -AgNO_3_ (99%), barium nitrate -Ba(NO_3_)2 (99%), zinc chloride -ZnCl_2_ (97.2%), ferric chloride (III) -FeCl_3_ (97%), ferrous chloride (II) -FeCl_2_ (98%), tin chloride -SnCl_2_ (99%), lead chloride -PbCl_2_ (99%), copper chloride -CuCl_2_ (99%), cobalt chloride -CoCl_2_ (98%), and mercury chloride -HgCl_2_ (99.5%) were obtained from sigma Aldrich.
Equipment used in this study consists of a UV–Vis (Perkin Elmer Lambda 950) spectrophotometer and fluorescence (Varian Cary Eclipse) spectrometer. FT-IR (Varian 660), Spectrofluorometer (Varian Cary Eclipse) was used to evaluate the presence of main functional groups present in CDs. The size distribution of CDs and crystallographic planes were acquired through TEM-JEOL JEM-2010 images. A Thermo Scientific scalab 250Xi, X-ray photoelectron spectrometer (XPS) was used to analyze the chemical and structural composition of CDs. The FE-SEM (Hitachi SU5000) equipment was employed to study the morphology and composition (Energy Dispersive X-ray spectroscopy) of heavy metal/CDs microparticles.
CDs fabrication
The CDs were prepared according to Torres Landa et al.75 Briefly, manila tamarind leaves were washed and dried in a dehydrator at 40°C for 4 h. The dried carbon precursor was ground in a coffee mill for 30 s. CDs were synthesized by direct carbonization in a muffle furnace at 200 (CDs_200_) and at 400°C (CDs_400_) for 120 min. The generated carbonaceous material was utilized to make suspensions with a concentration of 10 mg/mL that were treated in an ultrasonic bath for 4 h before being centrifuged at 13,000 rpm for 20 min and, filtered using 2 μm filter paper. The filtrate was refrigerated to preserve the CDs to be used for further studies. The relative quantum yield (QY) was calculated to be 24 % (CD_200_) using rhodamine 6G in water (QY = 92%).
Sensing and removal procedure
The specificity assessment was developed using stock sol^n^ of CDs and metal ions (10 mM) in a volumetric ratio of 1:119. The sensitivity study consists in the measurement of PL intensity with the addition of different concentration of metal ions using the same proportion of CDs and metal. The precipitation test was performed by setting the CDs:metal solution in a volume ratio of 1:5. The used concentration range of metal ions is 160 μM – 5.8 mM. The removal efficiency (%R) was calculated with the equation , ( C0 and C are the concentration of metal ions before and after adsorption). Dissolutions were centrifuged (13,000 rpm, 2 min) to obtain CDs/heavy metal sediments and dried and analyzed using XPS and FE-SEM.
Machine learning models
Different machine learning classification models were utilized to categorize the collected data,76 Support Vector Machines (SVMs) operate on the principle of a margin, which defines the space on either side of a hyperplane that separates two data classes. By maximizing this margin, SVMs ensure the greatest possible distinction between the hyperplane and the nearest data points from each class. K-Means clustering, an unsupervised learning technique, segments a dataset into a predetermined number of clusters. The algorithm iteratively assigns each data point to the closest cluster center and updates the centroids based on the average position of the assigned points. This process continues until the clusters stabilize, ensuring convergence. Decision Trees (DTs) are structured models that classify data instances by recursively dividing the dataset based on feature values. Each internal node represents a decision rule derived from a specific feature, each branch corresponds to a possible outcome, and the leaf nodes indicate final class labels. Due to their transparency and effectiveness in handling non-linear patterns, decision trees are widely employed in classification and regression tasks.
Quantification and statistical analysis
There are no quantification or statistical analyses to include in this study.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1Azam M.Khan M.R.Wabaidur S.M.Al-Resayes S.I.Islam M.S.Date pits waste as a solid phase extraction sorbent for the analysis of lead in wastewater and for use in manufacturing brick: An eco-friendly waste management approach J. Saudi Chem. Soc.26202210151910.1016/j.jscs.2022.101519 · doi ↗
- 2Liu J.Huang Z.Chen Z.Sun J.Gao Y.Wu E.Resource utilization of swine sludge to prepare modified biochar adsorbent for the efficient removal of Pb(II) from water J. Clean. Prod.257202012032210.1016/j.jclepro.2020.120322 · doi ↗
- 3Kong Y.Wang M.Lu W.Li L.Li J.Chen M.Wang Q.Qin G.Cao D.Rhodamine-based chemosensor for Sn 2+ detection and its application in nanofibrous film and bioimaging Anal. Bioanal. Chem.41420222009201910.1007/s 00216-021-03836-035048136 · doi ↗ · pubmed ↗
- 4Singh G.Singh K.N.Gonzalez-Silvera D.Espinosa-Ruiz C.Esteban M.A.Functionalized organosilanes and their magnetic nanoparticles as receptor for Sn (II) ions detection and potent antioxidants J. Mol. Struct.1247202213129710.1016/j.molstruc.2021.131297 · doi ↗
- 5Dong W.-K.Akogun S.F.Zhang Y.Sun Y.-X.Dong X.-Y.A reversible ‘turn-on’ fluorescent sensor for selective detection of Zn 2+Sens. Actuators B Chem.238201772373410.1016/j.snb.2016.07.047 · doi ↗
- 6Hussain S.Khan M.Sheikh T.M.M.Mumtaz M.Z.Chohan T.A.Shamim S.Liu Y.Zinc Essentiality, Toxicity, and Its Bacterial Bioremediation: A Comprehensive Insight Front. Microbiol.13202290074010.3389/fmicb.2022.900740 PMC 919758935711754 · doi ↗ · pubmed ↗
- 7Liu X.Chen Z.Gao R.Kan C.Xu J.A paper-based fluorescent and colorimetric portable device with smartphone application for Fe 3+ sensing J. Environ. Chem. Eng.10202210765010.1016/j.jece.2022.107650 · doi ↗
- 8Hsieh C.-T.Sung P.-Y.Gandomi Y.A.Khoo K.S.Chang J.-K.Microwave synthesis of boron- and nitrogen-codoped graphene quantum dots and their detection to pesticides and metal ions Chemosphere 318202313792610.1016/j.chemosphere.2023.13792636682636 · doi ↗ · pubmed ↗