Skip to main content
NCBI home page
ACS Omega logoLink to ACS Omega
. 2026 Apr 24;11(17):25505–25514. doi: 10.1021/acsomega.5c13597

Differentiation of Lupinus and Mimosa by Machine Learning Employing Spectroscopic Data: A Comparative Study

Mariana Koetz †, Maria H Vendruscolo †, Andressa K Maia †, Stefanie L K Martinelli †, Marina González ‡, Milton K Sasaki ‡, Sílvia T S Miotto §, Marina Scopel ∥, Elen de Oliveira Alves ⊥, Rodrigo M Viegas #, Alexandre M Fuentefria †, Marco F Ferrão ‡, José A S Zuanazzi †,*
PMCID: PMC13150596  PMID: 42110731

Abstract

This study investigated the complex chemical composition of species from the genera Lupinus and Mimosa (Fabaceae) from southern Brazil using ultraviolet–visible (UV–vis) and Fourier transform infrared (FTIR) spectroscopy, in conjunction with machine learning techniques. The results revealed that, despite similarities in the presence of secondary metabolites such as flavonoids and alkaloids, the genera can be accurately distinguished. To overcome the high dimensionality and collinearity of spectral data, linear discriminant analysis (LDA) was coupled with variable selection by successive projection algorithm (SPA) and genetic algorithm (GA). These models successfully isolated specific spectral markers, achieving 100% sensitivity, specificity, and accuracy in sample classification. This study highlights the effectiveness of combining accessible spectroscopic techniques with machine learning in plant taxonomy, offering a robust preliminary alternative for species differentiation and enhancing the understanding of chemical diversity within the Fabaceae family.

graphic file with name ao5c13597_0009.jpg

graphic file with name ao5c13597_0007.jpg

1. Introduction

The Fabaceae (Leguminosae) family is one of the largest among Angiosperms, comprising approximately 795 genera and almost 20,000 species, distributed throughout the world, mainly in subtropical and tropical regions. − This group stands out for its richness in bioactive compounds and its ecological and economic importance, notably including plants that establish symbiosis with nitrogen-fixing bacteria in the soil. −

Species of the Fabaceae family possess several compounds, including alkaloids, flavonoids, terpenoids, tannins, and saponins, which have been investigated for their pharmacological properties, such as anti-inflammatory, antioxidant, and antimicrobial activities. −

The genera Lupinus L. and Mimosa L. (Fabaceae) are notable for the complexity of their chemical profiles and potential applications of their bioactive compounds. Lupinus lanatus Benth. and Lupinus luteus L. are widely studied due to the presence of quinolizidine alkaloids. These alkaloids exhibit significant pharmacological properties, including analgesic, antiarrhythmic, and antitumor effects. , Furthermore, species of the genus Lupinus, owing to the ability of legumes to fix atmospheric nitrogen, are also used as sources of protein and play a fundamental role in sustainable agriculture, contributing to soil recovery. ,

Similarly, species of the genus Mimosa present diverse bioactivities and therapeutic potential. Martínez-Higuera et al. reported that a hydrogel with silver nanoparticles synthesized with the addition of Mimosa tenuiflora (Willd.) Poir. demonstrated bactericidal and anti-inflammatory effects, accelerating the recovery of burns. Valencia-Gómez et al. also observed that O-carboxymethyl chitosan films enriched with M. tenuiflora extract maintain their physicochemical properties and promote rapid healing by stimulating inflammation and fibroblast proliferation. Furthermore, studies with Mimosa pigra L. extracts have demonstrated the ability to inhibit the proliferation of WEHI-3 cells, contributing to the treatment of acute myeloid leukemia, while Nguyen et al. highlighted the pulmonary antifibrotic potential of this genus, protecting alveolar epithelial cells and fibroblasts against inflammatory injuries during lung fibrogenesis.

The immense chemical complexity and taxonomic diversity within the Fabaceae family demand rigorous analytical strategies to fully characterize their metabolite profiles, ensure quality control of their bioactive extracts, and resolve taxonomic ambiguities. To achieve these scientific objectives, researchers have increasingly relied on comprehensive chemical fingerprinting and metabolomics. Consequently, several species of this family have been extensively studied using a variety of advanced analytical techniquessuch as NMR, flame atomic absorption spectrometry (F-AAS), GC–MS, LC–MS, HPLC–MS/MS, UPLC-QTOF-MS, and capillary electrophoresisfor the classification of chemical profiles and data processing using chemometric methods. For instance, high-performance thin-layer chromatography was used to compare the fingerprint of propolis samples with those of plant bud extracts and, through palynological analysis, identified that the pollen grains of the propolis samples belonged to the families Asteraceae, Fabaceae, Rosaceae, among others. HPLC was used for chemometric studies and phenolic metabolomics of Cassia absus seeds L. F-AAS was used to classify the elemental profiles of legumes and oilseeds. The contents of oligosaccharides of the raffinose and sucrose families in Brassica L., Lupinus, Pisum L., and Hordeum L. species were investigated by chemometric principal component analysis (PCA) using capillary electrophoresis. Furthermore, GC–MS, LC–MS, and 1D NMR techniques were used to compare the metabolite profile and fingerprint of 25 compounds from four species (Glycyrrhiza echinata L., Glycyrrhiza glabra L., Glycyrrhiza inflata Batalin, and Glycyrrhiza uralensis Fisch ex DC.). The species Sutherlandia frutescens (L.) R. Br. and Sutherlandia microphylla Burch ex DC. were compared to establish phytochemical variations and chemometric analysis using UPLC-TQ-MS/MS.

Although these advanced analytical methods are capable of providing detailed chemical profiles and accurate structural information, which enable the identification and quantification of a wide variety of bioactive compounds, they have the disadvantages of requiring expensive equipment, specialized maintenance, and trained operators, thereby limiting the accessibility of these techniques and rendering them unfeasible for laboratories with more restricted resources. −

To overcome the limitations of high-cost equipment and provide a scalable alternative for species differentiation, this study proposes the use of Fourier transform infrared (FTIR) and ultraviolet–visible (UV–vis) spectroscopy. The fundamental hypothesis motivating the choice of these complementary techniques is that the distinct secondary metabolite profiles of Lupinus (e.g., quinolizidine alkaloids) and Mimosa (e.g., diverse flavonoids and tannins) will produce unique and discriminative spectral signatures. Specifically, UV–vis spectroscopy is highly sensitive to conjugated π-systems, making it an excellent tool for detecting variations in phenolic compounds and flavonoids. Concurrently, FTIR spectroscopy provides a comprehensive vibrational fingerprint of molecular functional groups, such as carbonyls and aromatic rings, which are essential for identifying alkaloid and phenolic backbones. Previous studies have demonstrated that these techniques, in addition to being simpler and highly accessible, exhibit good reproducibility and provide robust chemical information. , The possibility of associating these specific spectral signatures with machine learning algorithms allows the generated data to be used to build robust predictive and classification models, , enabling the identification of underlying chemical patterns and facilitating the reliable separation of different genera and species. ,

Chemometric methods encompass machine learning techniques that enable the analysis and interpretation of complex chemical data and are divided into supervised and unsupervised methods, each with specific applications aimed at optimizing information extraction. ,

Among the unsupervised methods, we highlight PCA and hierarchical cluster analysis (HCA). These methods are particularly useful for identifying patterns in data sets. PCA reduces the dimensionality of the data, highlighting the main patterns and variables while preserving as much of the original information as possible. HCA groups samples are based on similarities and are used for classification purposes.

In complex plant matrices, such as those of the Fabaceae family, spectral data often exhibit high dimensionality and severe band overlapping. While conventional exploratory chemometrics (like PCA) can reveal data structure, they are often insufficient for identifying the specific, nonobvious spectral features responsible for species differentiation. Machine learning algorithms, particularly those coupled with variable selection like successive projection algorithm (SPA) and genetic algorithm (GA), overcome this by isolating the most discriminant spectral variables, eliminating noise and redundancy, and providing deeper insights into the specific chemical markers driving the classification. ,,

Supervised methods, which require prior information about sample classes, include linear discriminant analysis (LDA), in which the SPA and GA are prominent. SPA is used to efficiently select relevant variables, minimizing redundancy and improving the accuracy of predictive models, while GA, inspired by biological evolution, randomly selects new individuals in the current population at each repetition. , The application of these methods to natural product data facilitates the classification and discrimination of bioactive compounds, aiding in the differentiation between species and understanding their chemical and biological properties.

Given the above, this study aims to investigate the genera Lupinus and Mimosa using UV/vis and FTIR techniques in conjunction with machine learning in order to develop predictive models based on chemical data from samples collected in Rio Grande do Sul (Brazil), effectively classifying and differentiating the genera.

2. Materials and Methods

2.1. Reagents and Materials

The following chemical reagents were used to obtain the extracts: methanol and ethanol (Neon, Suzano, SP, Brazil); ethyl ether (Dinâmica, Indaiatuba, SP, Brazil); and distilled water (GFL 2008 Water Distiller).

2.2. Plant Material

Samples of Lupinus (n = 15) and Mimosa (n = 18) were collected in Rio Grande do Sul, Brazil (Table ), identified, and their voucher specimens were deposited in the ICN-UFRGS Herbarium. The samples were dried and ground in a knife mill, and their particle size was standardized to 500–710 μm.

1. Set of Plant Species Collected in Rio Grande do Sul, Brazil.

Mimosa
Lupinus
code species collection number code species collection number
MI01 sp1 205,639 LU01 albescens 205,647
MI02 sp3 205,672 LU02 bracteolaris 205,608
MI03 sp4 205,673 LU03 bracteolaris 205,611
MI04 dolens 205,862 LU04 bracteolaris 205,636
MI05 bifurca 205,675 LU05 bracteolaris 205,652
MI06 baptistae 205,650 LU06 bracteolaris 205,660
MI07 glycyrrhizoides 205,858 LU07 gibertianus 205,668
MI08 ramurosa 205,635 LU08 guaraniticus 205,872
MI09 dutrae 205,670 LU09 lanatus 205,609
MI10 cruenta 205,642 LU10 lanatus 205,644
MI11 daleoides 205,850 LU11 lanatus 205,654
MI12 myriophylla 205,884 LU12 multiflorus 205,630
MI13 pilulifera 205,640 LU13 paranensis 205,631
MI14 riverensis 205,847 LU14 paranensis 205,633
MI15 schleidenii 205,665 LU15 multiflorus 205,883
MI16 bimucronata 210,936      
MI17 pigra 210,935      
MI18 pudica –      
Open in a new tab
a

Exsiccata not yet deposited.

2.3. Preparation of Lyophilized Extracts

The metabolites were extracted with analytical grade methanol at a drug-to-solvent ratio of 1:20 using four 20 min cycles of ultrasonic extraction. The resulting solution was filtered, and the filtrate was collected in a round-bottom flask. The solvent was removed by evaporation using a rotary evaporator. The dried extract was resuspended in 30 mL of water, and chlorophyll was removed by four cycles of washing with 20 mL of diethyl ether. The aqueous phase was frozen and lyophilized. The lyophilized extracts were used in the experiments.

2.4. Ultraviolet–Visible (UV–Vis) Spectrophotometry Analysis

Lyophilized extracts were prepared as solutions of 10 μg/mL in 80% ethanol. These solutions were prepared in triplicate and analyzed using an Agilent Technologies 8453 UV–vis spectrophotometer in scanning mode (240–400 nm). The resulting visible spectral data were exported as a CSV file, and the numerical data matrix was used for multivariate analyses.

2.5. Fourier Transform Infrared Spectroscopy (FTIR) Analysis

Lyophilized extracts were analyzed in duplicate using an Agilent Technologies Cary 630 FTIR spectrometer with a zinc selenide (ZnSe) crystal and helium/neon laser. Spectra were generated in the region of 1800–600 cm–1, with 32 scans and a resolution of 4 cm–1. The resulting numerical data matrix was used for multivariate analyses.

2.6. Multivariate Analysis

2.6.1. Software

ChemoStat V2 was used for the HCA and PCA. MATLAB software, version 2014a (The MathWorks Inc., MA, USA), was used for LDA classification models, employing variable selection with GA and SPA. These models were developed using VS TOOLBOX, a proprietary toolbox software developed by the Analytical Instrumentation and Chemometrics group.

2.6.2. Unsupervised Methods

Exploratory analyses using HCA and PCA were applied to the UV–vis and FTIR data to reveal similarities between the samples. Data preprocessing is a critical step in minimizing instrumental noise and scattering effects before multivariate analysis. In this study, preprocessing included Savitzky–Golay smoothing (second-order polynomial, window size = 13 points), which was empirically selected to adequately remove high-frequency instrumental noise without oversmoothing or distorting the relevant spectral absorption bands. Subsequently, baseline correction and normalization were performed using min–max normalization (range 0–1) applied across the full spectral region of each individual sample. Finally, the data were globally mean-centered across the entire data set to ensure that all variables were evaluated based on their variance relative to the overall mean. Regarding data integrity, a rigorous visual and exploratory inspection of the spectral data and PCA scores was established to assess the presence of potential outliers prior to model training.

2.6.3. Supervised Methods

The direct application of classical LDA to full-spectrum data is generally unfeasible due to the high degree of collinearity and the fact that the number of variables greatly exceeds the number of samples, which inevitably leads to model overfitting and matrix singularity. To overcome this limitation and construct chemically interpretable models, LDA was coupled with variable selection algorithms. Two distinct approaches were tested to find the most parsimonious and mechanistically meaningful subset of variables: the SPA, a deterministic method that minimizes collinearity, and the GA, a stochastic evolutionary approach. Both algorithms aim not just to classify the samples but to isolate the most discriminant wavelengths/wavenumbers, providing a chemically informed explanation for the differentiation between the genera. The SPA was tested with variable selection windows (1–10) and GA with three tested combinations (population/generation: 100/100; 200/200; 300/300), 5% mutation, and 60% crossover. In each GA run, a maximum of 10 variables were selected. The Kennard–Stone method was used to divide the samples; 22 samples were selected for the training set (10 Lupinus and 12 Mimosa) and 11 samples for the test set (5 Lupinus and 6 Mimosa). The performance metrics were calculated using the training and test sets with the results for the training set being presented using cross-validation. To evaluate the LDA models, the following figures of merit were calculated: sensitivity (SEN) (eq ), specificity (SPE) (eq ), and accuracy (ACC) (eq ), where TP means true positive, TN true negative, FP false positive, and FN false negative.

Sensitivity(SEN)(%)=(TP/(TP+FN))×100 1
Specificity(SPE)(%)=(TN/(TN+FP))×100 2
Accuracy(ACC)(%)=((TP+TN)/(TP+FN+TN+FP))×100 3

where TP is the number of true positives (class A samples correctly classified as class A), TN is the number of true negatives (class B samples correctly classified as class B), FP is the number of false positives (class B samples incorrectly classified as class A), and FN is the number of false negatives (class A samples incorrectly classified as class B).

3. Results and Discussion

The genera Lupinus and Mimosa, despite belonging to the same family (Fabaceae), are taxonomically and evolutionarily distant. Lupinus belongs to the subfamily Papilionoideae, the tribe Genisteae. Mimosa currently belongs to the subfamily Caesalpinioideae, tribe Mimoseae, Mimosa clade. Given their placement in different subfamilies and their distinct morphologies (both vegetative and reproductive), differences in the presence and structure of their secondary metabolites are expected. Previous studies on M. tenuiflora, Mimosa artemisiana Heringer and Paula, Lupinus albus L. and Lupinus angustifolius L., and L. lanatus have reported a diversity of secondary metabolites, including flavonoids (flavones, isoflavones, flavonols, flavanones, and isoflavanones) and alkaloids (indolic and quinolizidine), among others.

Our findings in this study corroborate reports in the literature, supporting the differentiation of the two genera based on their chemical composition.

3.1. Multivariate Analysis

The UV–vis and FTIR spectral profiles of the average values for the samples of the two genera studied (Lupinus and Mimosa) after pretreatments are shown in Figure . By visually examining the spectra obtained from this exploratory analysis, we already observe similarities and differences between the profiles of the different samples analyzed. During this initial exploratory and visual inspection of the spectral data (Figure ) and subsequent PCA scores, no samples were identified as anomalous outliers. Consequently, no samples were excluded from the data set prior to model training. All samples were retained to accurately capture and reflect the natural biological and intraspecific variability inherent to the Lupinus and Mimosa species evaluated, thereby ensuring the development of robust and realistic predictive models.

1.

1

Open in a new tab

Average profiles obtained in the analyses by (A) UV–vis (240–400 nm) and (B) FTIR (1600–600 cm–1). Results obtained by the average of the analysis of all samples of each studied genera: LU: Lupinus and MI: Mimosa. Shaded areas represent the standard deviation (n = 100 for UV; n = 66 for FTIR).

3.1.1. Unsupervised Methods

The HCA and PCA results allowed for the identification of two distinct groups. The dendrograms (Figure ), scores and loadings of the principal components (Figures and ), support this observation.

2.

2

Open in a new tab

HCA-dendrogram (A): UV–vis and (B): FTIR. Results expressed by the average of the analysis of each sample, from the total set of the two genera: LULupinus and MIMimosa.

3.

3

Open in a new tab

PCA-UV–vis (240–400 nm). (A) Scores and (B) loadings. Results expressed by the average of the analysis of each sample, from the total set of the two genera: LULupinus and MIMimosa.

4.

4

Open in a new tab

PCA-FTIR (1800–600 cm–1). (A) Scores and (B) loadings. Results expressed by the average of the analysis of each sample, from the total set of the two genera: LULupinus and MIMimosa.

The principal components, PC1 and PC2, in both methods were able to explain and enable the visualization of strong patterns and relationships within the samples of the two studied genera. In the UV–vis analyses (Figure ), PC1 was able to explain 86.75% of the variability, while PC2 explained 7.72%. For FTIR (Figure ), these results were 48.28% (PC1) and 23.98% (PC2). By analyzing the PCA loading graphs, it is possible to observe the contribution of each original variable (wavelengths for UV–vis and wavenumbers for FTIR) to the PCs. There is a clear separation between the “LU” (blue) and “MI” (red) groups along PC1, suggesting that the original variables that contribute most to PC1 are those that best discriminate the two groups.

As shown in Figure , in PC1, the Lupinus samples are characterized by a predominance of positive signals, with more intense signals in the approximate regions between 259 and 269 and 340–350 nm. In contrast, the Mimosa samples show a predominance of negative signals, with more intense signals in the approximate region of 240 nm. In PC2, both genera exhibit both positive and negative signals.

Substitutions in the basic structure of flavonoids, such as hydroxylations or glycosylations, alter the wavelengths of maximum absorption in the UV–vis. Studies demonstrate that the number and position of hydroxyl groups in rings A and B modulate the absorption bands, ranging from 300 to 385 nm (Band II, Ring A) and 240–285 nm (Band I, Ring B). ,

Similarly, Figure shows that in PC1, Lupinus samples exhibit positive signals (with more intense signals in the approximate regions of 1580–1560 cm–1), and Mimosa samples exhibit negative signals (with more intense signals in the approximate regions of 1215–1200 cm–1). In PC2, positive signals are more related to Mimosa (with more intense signals around 1200 cm–1), while negative signals are more related to Lupinus (with more intense signals in the approximate regions of 1650, 1570, and 1410 cm–1).

FTIR spectroscopy reveals diagnostic wavenumber bands for identifying functional groups present in the structure of flavonoids. The regions between 1800–1540 cm–1, 1600–1300 cm–1, and 909–650 cm–1 correspond, respectively, to carbonyl groups (CO), aromatic rings, and aliphatic structures. The region of 1300–909 cm–1, although complex and characteristic of each molecule, is mainly used for spectral comparison and identity confirmation.

Because they have different chemical structures in their composition but belong to the same class of secondary metabolites (e.g., flavonoids and alkaloids), unsupervised exploratory analysis revealed a clear tendency for the two studied genera to separate into distinct groupings. This analysis also allowed for the visualization of wavelength and wavenumber regions found in the reported structures of both genera. Therefore, the natural separation observed in the PCA score plots is fundamentally driven by these specific spectral features: the variance in the UV–vis absorption bands (associated with distinct flavonoid substitutions) and the FTIR vibrational modes (especially the characteristic CO stretching of quinolizidine alkaloids in Lupinus and the C–O vibrations of specific flavonoids in Mimosa). This demonstrates that the discrimination is not merely a mathematical artifact but a direct reflection of their underlying chemical differences.

Based on the visualization and understanding of the data structure obtained through exploratory analysis (HCA and PCA), we performed a LDA to test whether these groups are significantly distinct.

3.1.2. Supervised Methods

3.1.2.1. Linear Discriminant Analysis (LDA): Successive Progression Algorithm (SPA) and Genetic Algorithm (GA)

Analysis of the averaged UV–vis (Figure A) and FTIR (Figure B) spectra reveals similarities between the genera Lupinus and Mimosa, which aligns with the literature describing flavonoids and alkaloids as the main metabolites present in both. , Despite the similarities, subtle differences in absorption patterns, such as peak shifts and variations in band intensity, indicate variations in the chemical composition between the genera.

Analyses with variable selection of spectral data (SPA-LDA and GA-LDA) allowed the identification of distinct groups corresponding to genera Lupinus and Mimosa, demonstrating excellent performance in sample classification.

The results of the figures of merit, calculated from eqs –, to test the LDA models were all satisfactory, with values between 95 and 100% for all figures (sensitivity, specificity, and accuracy).

The best results obtained by SPA-LDA and GA-LDA for the UV–vis data are presented in Figure . The best result for GA-LDA was obtained with the following combination: population: 100; generation: 100; mutation 5%; crossover 60%. Figure A,B shows the variables selected for each algorithm tested, with one variable in SPA and four variables in GA, respectively. The Fisher weight plots (Figure C,D) demonstrate the contribution of the selected variables to each genus studied (256 nm in SPA; 257 and 356 nm in GA, related to Lupinus, and 284 and 261 nm in GA, related to Mimosa). The discriminant function graphs (Figure E,F) demonstrate that the samples were categorized with an excellent correct classification rate, with 100% in the training (circles) and test (stars) sets for the two algorithms tested.

5.

5

Open in a new tab

UV–vis results: SPA-LDA (A,C,E); GA-LDA (B,D,F). (A,B) Variable selection in the two tested algorithms; (C,D) Fisher weights with wavelengths selected in the two tested algorithms; and (E,F) training sets (circles) and test sets (stars) with the separation of the two analyzed classes, in the two tested algorithms. LU: Lupinus and MI: Mimosa.

Figure shows the best results obtained by SPA-LDA and GA-LDA for the FTIR data. The best result for GA-LDA was obtained with the combination: population: 100; generation: 100; mutation 5%; and crossover 60%. Figure A,B shows the variables selected for each algorithm tested with six variables in SPA (Figure A) and nine variables in GA (Figure B). The samples were also categorized with an excellent correct classification rate, with 100% in the training (circles) and test (stars) sets for both algorithms tested (Figure E,F). Finally, the contribution of the selected variables to each genus studied can be observed in Figure C,D (1658, 1595, and 665 cm–1 in SPA; 1476, 1105, 828, 791, and 660 cm–1 in GA, related to Lupinus, and 1561, 1185, and 1103 cm–1 in SPA; and 1535, 1107, 781, and 675 cm–1 in GA, related to Mimosa).

6.

6

Open in a new tab

FTIR results: SPA-LDA (A,C,E); GA-LDA (B,D,F). (A,B) Variable selection in the two tested algorithms; (C,D) Fisher weights with wavenumber selected in the two tested algorithms; and (E,F) training sets (circles) and test sets (stars) with the separation of the two analyzed classes, in the two tested algorithms. LU: Lupinus and MI: Mimosa.

Despite the excellent classification metrics achieved, it is important to briefly outline the applicability domain of the proposed models. The current predictive models were developed and validated using species collected specifically in the Rio Grande do Sul region (Table ). Since plant secondary metabolism can be influenced by geographic, climatic, and seasonal variations, the direct application of these specific models to Lupinus and Mimosa species from entirely different biomes may require prior model updating or recalibration. Furthermore, although the models successfully discriminate between these two genera, their applicability does not currently extend to identifying individual species within the genera, which remains a limitation to be addressed in future studies with larger sample sizes.

The similarity in wavelengths (Figures A and A) and wavenumbers (Figures B and B) selected by the algorithms to classify/separate the genera Lupinus and Mimosa reflects the presence of secondary metabolites from common classes but with structural differences.

To provide a chemical rationale for the discrimination achieved by the LDA models, the most significant variables selected by SPA and GA were tentatively assigned to specific functional groups and secondary metabolites known to occur in these genera. This correlation confirms that the mathematical models are grounded in the distinct biosynthetic pathways of each genus. Table summarizes these spectral assignments, linking the mathematical markers to the underlying chemical profiles of Lupinus (notably quinolizidine alkaloids) and Mimosa (rich in diverse flavonoids).

2. Tentative Assignment of Key Spectral Variables Selected by SPA-LDA and GA-LDA for the Differentiation of Lupinus and Mimosa Genera.
technique selected variable tentative assignment chemical relevance refs
UV–vis ∼259 nm Band II (Ring A) absorption of isoflavones presence of genistein and derivatives in Lupinus ,
UV–vis 284 nm Band II absorption/phenolic B-ring variation in hydroxyl/methoxyl substitutions ,
UV–vis 356–361 nm Band I (Ring B) absorption of flavonols flavones and flavonols (e.g., quercetin) in Mimosa ,,
FTIR 1658 cm–1 CO stretching (Amide I) carbonyl groups in quinolizidine alkaloids ,
FTIR 1103–1107 cm–1 C–Ostretching vibrations flavonoid glycosides and phenolic backbones
Open in a new tab

Previous studies have identified acylated and glycosylated derivatives of myricetin and quercetin in M. pigra, while M. artemisiana revealed the presence of these flavones, in addition to hydroxylated and dimethoxylated derivatives of kaempferol. The aglycones of these flavonols exhibit maximum absorption bands in the 360 nm region, attributed to ring A, with variations resulting from substitutions in this ring. This signal, therefore, suggests that it is a significant marker for the Mimosa genus. Additionally, the detection of gallic acid and catechin in Mimosa pudica and M. tenuiflora, , compounds with maximum absorption near 280 nm (Ring B), reinforces the importance of the 284 nm variable (lowest intensity) in the discrimination between the genera. The variations observed around 280 nm are associated with substitutions by hydroxyl, methoxyl, or glycoside groups. ,

Findings from other studies reported the presence of the isoflavone genistein in L. albus and L. angustifolius, with a maximum absorption wavelength at 259 nm. Additionally, quinolizidine alkaloids, identified in L. lanatus, also showed absorption in this spectral region. These findings corroborate the results of the present study and emphasize the relevance of the selected variable for the classification of genera Lupinus in relation to genus Mimosa.

Fisher weight plot analysis of FTIR spectra (Figure C,D) revealed signals suggestive of clusters relevant to the discrimination between genera Lupinus and Mimosa. For Lupinus, the wavenumber 1658 cm–1 (CO) was selected as significant, which is strongly suggestive of the presence of carbonyls typically found in quinolizidine alkaloids, which is characteristic of the genus. In Mimosa, the wavenumbers 1103 and 1107 cm–1 (C–O), observed in SPA and GA, respectively, are tentatively assigned to stretching vibrations of C–O bonds, which is consistent with the abundance of flavonoids reported for this genus.

Despite the presence of similar classes of secondary metabolites, which explains the proximity of UV–vis absorption wavelengths and FTIR chemical shifts for both genera included in this study, the combination of simple analytical techniques and chemometrics allowed the correct classification of all species. The high accuracy and sensitivity of the applied chemometric methods (LDA with successive progressions and GAs) for both analytical methods (UV–vis and FTIR) demonstrate the effectiveness of the approach. Cross-validation and performance metrics confirm the robustness of the models, validating LDA as an effective tool for complex data analysis without the need for advanced techniques such as NMR or chromatography.

4. Conclusion

Our study demonstrated the effectiveness of UV–vis and FTIR spectroscopy, combined with variable selection algorithms, in distinguishing species from the genera Lupinus and Mimosa collected in Rio Grande do Sul, Brazil. Subtle spectral variations, which are tentatively assigned to differences in specific secondary metabolites (such as quinolizidine alkaloids and flavonoids), allowed accurate discrimination. The coupling of LDA with the SPA and GA proved to be essential to overcome data collinearity and identify mechanistically meaningful spectral markers, yielding highly accurate predictive models. While future studies encompassing broader geographic regions and larger sample sizes are necessary to expand the applicability domain, this research highlights the potential of machine learning-assisted spectroscopy as a robust auxiliary tool in plant taxonomy, complementing traditional morphological analyses.

Acknowledgments

Instituto Nacional de Ciência e Tecnologia em Biodiversidade e Produtos NaturaisINCT BioNat, grant number 465637/2014-0, Brazil.

This work was supported by the Conselho Nacional de Desenvolvimento Científico e TecnológicoProject number 424317/2018-4. (CNPqgrant number: 309,035/2022-8). J.A.S.Z. received the CNPq Research Productivity Grant.

The Article Processing Charge for the publication of this research was funded by the Coordenacao de Aperfeicoamento de Pessoal de Nivel Superior (CAPES), Brazil (ROR identifier: 00x0ma614).

The authors declare no competing financial interest.

During the preparation of this work, the author used the AI tool Gemini 2.5 Flash to improve the writing and language of the manuscript. After using this tool, the author reviewed and edited the content as necessary and assumes full responsibility for the content of the published article.

References

  1. Christenhusz M. J. M., Byng J. W.. The number of known plant species in the world and its annual increase. Phytotaxa. 2016;261(3):201–217. doi: 10.11646/phytotaxa.261.3.1. [DOI] [Google Scholar]
  2. Azani N.. A new subfamily classification of the Leguminosae based on taxonomically comprehensive phylogeny: The Legume Phylogeny Working Group (LPWG) Taxon. 2017;66(1):44–77. doi: 10.12705/661. [DOI] [Google Scholar]
  3. Reflora . Fabaceae in Flora and Funga of Brazil. Rio de Janeiro Botanical Garden. Available at: https://floradobrasil.jbrj.gov.br/FB115 (accessed Nov 16, 2024).
  4. Peix A., Ramírez-Bahena M. H., Velázquez E., Bedmar E. J.. Bacterial Associations with Legumes. Crit. Rev. Plant Sci. 2015;34:17–42. doi: 10.1080/07352689.2014.897899. [DOI] [Google Scholar]
  5. Sprent J. I., Ardley J., James E. K.. Biogeography of nodulated legumes and their nitrogen-fixing symbionts. New Phytol. 2017;215:40–56. doi: 10.1111/nph.14474. [DOI] [PubMed] [Google Scholar]
  6. Schwember A. R., Schulze J., del Pozo A., Cabeza R. A.. Regulation of Symbiotic Nitrogen Fixation in Legume Root Nodules. Plants. 2019;8:333. doi: 10.3390/plants8090333. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Neves A. M., Costa P. S., Coutinho M. G. S., Souza E. B., Santos H. S., Silva M. G. V., Fontenelle R. O. S.. Chemical Characterization and the Antimicrobial Potential of Species of the Genus Senna Mill. (Fabaceae) Rev. Virtual Quim. 2017;9(6):2506–2538. doi: 10.21577/1984-6835.20170149. [DOI] [Google Scholar]
  8. Morais C. B., ScopeL M., Pedrazza G. P. R., Silva F. K., Dalla Lana D. F., Tonello M. L., Miotto S. T. S., Machado M. M., Oliveira L. F. S., Fuentefria A. M., Zuanazzi J. A. S.. Anti-dermatophyte activity of Leguminosae plants from Southern Brazil with emphasis on Mimosa pigra (Leguminosae) Mycol. J. Méd. 2017;27:530–538. doi: 10.1016/j.mycmed.2017.07.006. [DOI] [PubMed] [Google Scholar]
  9. Chen L., Tang X., Yang Q., Cheng X.. Quantitative and Chemical Fingerprint Analysis of Desmodium styracifolium by High-Performance Liquid Chromatography Combined with Chemometrics. J. Chromatogr. Sci. 2020;58(4):294–302. doi: 10.1093/chromsci/bmz112. [DOI] [PubMed] [Google Scholar]
  10. Avula B., Bae J. Y., Chittiboyina A. G., Wang Y. H., Wang M., Zhao J., Ali Z., Brinckmann J. A., Li J., Wu C., Khan I. A.. Chemometric analysis and chemical characterization for the botanical identification of Glycyrrhiza species (G. glabra, G. uralensis, G. inflata, G. echinata and G. lepidota) using liquid chromatography-quadrupole time of flight mass spectrometry (LC-QToF) J. Food Compos. Anal. 2022;112:104679. doi: 10.1016/j.jfca.2022.104679. [DOI] [Google Scholar]
  11. Rha C. S., Jang E. K., Lee J. S., Kim J. S., Ko M. J., Lim S., Park G. H., Kim D. O.. Statistical discrimination using different machine learning models reveals dissimilar key compounds of soybean leaves in targeted polyphenol-metric metabolomics in terms of traits and cultivation. Food Chem. 2023;404(A):134454. doi: 10.1016/j.foodchem.2022.134454. [DOI] [PubMed] [Google Scholar]
  12. Oladoja F. A., Awodele O., Oreagba I. A., Irokosu E. S., Oyinloye E. O., Murtala A. A.. Anti-inflammatory activity and molecular docking studies of the hydromethanolic leaf extract of Baphia longipedicellata brumitt in rats. Pharmacol. Res.Mod. Chin. Med. 2024;13:100512. doi: 10.1016/j.prmcm.2024.100512. [DOI] [Google Scholar]
  13. Neto A. T., Oliveira C. Q., Ilha V., Pedroso M., Burrow R. A., Dalcol I. I., Morel A. F.. Quinolizidine alkaloids from Lupinus lanatus . J. Mol. Struct. 2011;1004(1–3):174–177. doi: 10.1016/j.molstruc.2011.07.061. [DOI] [Google Scholar]
  14. Ishaq A. R., El-Nashar H. A. S., Younis T., Mangat M. A., Shahzadi M., Ul Haq A. S., El-Shazly M.. Genus Lupinus (Fabaceae): a review of ethnobotanical, phytochemical and biological studies. J. Pharm. Pharmacol. 2022;74(12):1700–1717. doi: 10.1093/jpp/rgac058. [DOI] [PubMed] [Google Scholar]
  15. Coba de la Peña T., Pueyo J. J.. Legumes in the reclamation of marginal soils, from cultivar and inoculant selection to transgenic approaches. Agron. Sustainable Dev. 2012;32:65–91. doi: 10.1007/s13593-011-0024-2. [DOI] [Google Scholar]
  16. Ortúzar M., Riesco R., Criado M., Alonso M. P., Trujillo M. E.. Unraveling the dynamic interplay of microbial communities associated to Lupinus angustifolius in response to environmental and cultivation conditions. Sci. Total Environ. 2024;946:174277. doi: 10.1016/j.scitotenv.2024.174277. [DOI] [PubMed] [Google Scholar]
  17. Martínez-Higuera A., Rodríguez-Beas C., Villalobos-Noriega J. M. A., Arizmendi-Grijalva A., Ochoa-Sánchez C., Larios-Rodríguez E., Martínez-Soto J. M., Rodríguez-León E., Ibarra-Zazueta C., Mora-Monroy R.. et al. Hydrogel with silver nanoparticles synthesized by Mimosa tenuiflora for second-degree burns treatment. Sci. Rep. 2021;11:11312. doi: 10.1038/s41598-021-90763-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Valencia-Gómez L. E., Martel-Estrada S. A., Vargas-Requena C. L., Acevedo-Fernández J. J., Rodríguez-González C. A., Hernández-Paz J. F., Santos-Rodríguez E., Olivas-Armendáriz I.. Characterization and evaluation of a novel O-carboxymethyl chitosan films with Mimosa tenuiflora extract for skin regeneration and wound healing. J. Bioact. Compat. Polym. 2020;35(1):39–56. doi: 10.1177/0883911519885976. [DOI] [Google Scholar]
  19. Deng W., Li H., Zeng Y., Wang Q.. Effect of ethyl acetate extract from Mimosa pudica root on acute myeloid leukemia in mice. J. China Pharm. 2024;35(11):1315–1319. doi: 10.6039/j.issn.1001-0408.2024.11.06. [DOI] [Google Scholar]
  20. Nguyen Q. C., Nguyen H. A., Pham T. A., Tran V. T.-H., Nguyen T. D., Pham D. V.. Mimosa pudica L. extract improves pulmonary fibrosis via modulation of MAPK signaling pathways and FOXO3 stabilization. J. Ethnopharmacol. 2024;330:118226. doi: 10.1016/j.jep.2024.118226. [DOI] [PubMed] [Google Scholar]
  21. Guzelmeric E., Ristivojević P., Trifković J., Dastan T., Yilmaz O., Cengiz O., Yesilada E.. Authentication of Turkish propolis through HPTLC fingerprints combined with multivariate analysis and palynological data and their comparative antioxidant activity. LWT. 2018;87:23–32. doi: 10.1016/j.lwt.2017.08.060. [DOI] [Google Scholar]
  22. Kaur P., Misser P., Kumar V., Singh R.. HPLC-MS based metabolomics & chemometrics studies of Cassia absus seeds extracted through green extraction technology. Phytomedicine Plus. 2024;4(3):100591. doi: 10.1016/j.phyplu.2024.100591. [DOI] [Google Scholar]
  23. Grembecka M., Szefer P.. Elemental Profiles of Legumes and Seeds in View of Chemometric Approach. Appl. Sci. 2022;12:1577. doi: 10.3390/app12031577. [DOI] [Google Scholar]
  24. Andersen K. E., Bjergegaard C., Møller P., Sørensen J. C., Sørensen H.. Compositional Variations for α-Galactosides in Different Species of Leguminosae, Brassicaceae, and Barley: A Chemotaxonomic Study Based on Chemometrics and High-Performance Capillary Electrophoresis. J. Agric. Food Chem. 2005;53(14):5809–5817. doi: 10.1021/jf040471v. [DOI] [PubMed] [Google Scholar]
  25. Farag M. A., Porzel A., Wessjohann L. A.. Comparative metabolite profiling and fingerprinting of medicinal licorice roots using a multiplex approach of GC–MS, LC–MS and 1D NMR techniques. Phytochemistry. 2012;76:60–72. doi: 10.1016/j.phytochem.2011.12.010. [DOI] [PubMed] [Google Scholar]
  26. Acharya D., Enslin G., Chen W., Sandasi M., Mavimbela T., Viljoen A.. A chemometric approach to the quality control of Sutherlandia (cancer bush) Biochem. System. Ecol. 2014;56:221–230. doi: 10.1016/j.bse.2014.06.009. [DOI] [Google Scholar]
  27. Wang X., Sun H., Zhang A., Wang P., Han Y.. Ultra-performance liquid chromatography coupled to mass spectrometry as a sensitive and powerful technology for metabolomic studies. J. Sep. Sci. 2011;34:3451–3459. doi: 10.1002/jssc.201100333. [DOI] [PubMed] [Google Scholar]
  28. Gonçalves E. S., Borges R. M., Carvalho L. V. B., Alves S. R., André L. C., Moreira J. C.. Analytical strategies with chromatography and mass spectrometry for biomonitoring of benzene exposure by determination of urinary S-phenylmercaturic acid. Braz. J. Occup. Health. 2017;42(suppl 1):e1s. doi: 10.1590/2317-6369000127615. [DOI] [Google Scholar]
  29. Ursulino J. S., Crispim A. C., Freire L. M., Matos E. R. C. S., Peçanha de Miranda Coelho J. A., Neto M. A. L., Ferreira da Silva Júnior E., Mendonça de Aquino T.. NMR-Based Metabolomics: General Aspects and Applications in Cancer Diagnosis. Appl. NMR Spectrosc. 2021;9:160–203. doi: 10.2174/9789815039351121090007. [DOI] [Google Scholar]
  30. González M., Mariotti K. C., Gomes A. A., Ferrão M. F., Limberger R. P.. Fingermark Analysis by Fourier Transform Infrared Microscopy Using Chemometric Tools. Braz. J. Anal. Chem. 2021;8(32):141–154. doi: 10.30744/brjac.2179-3425.AR-25-2021. [DOI] [Google Scholar]
  31. Wang Y., Tong J., Li X., Han X., Qin Y., Fang R., Gao M.. Attenuated Total Reflection Fourier Transform Infrared Spectral Identification of Bioaerosol Based on 1D-CNN. Acta Opt. Sin. 2024;44(24):2430003. doi: 10.3788/AOS231963. [DOI] [Google Scholar]
  32. Junges C. H., War C. C., Reis N. A. P. C., Gomes A. A., Diogo F. S., Ferrão M. F.. Grape juice classification with respective agricultural production system by means of visible spectroscopy assisted chemometrics. J. Food Compos. Anal. 2022;114:104793. doi: 10.1016/j.jfca.2022.104793. [DOI] [Google Scholar]
  33. Junges C. H., Guerra C. C., Gomes A. A., Ferrão M. F.. Green analytical methodology for grape juice classification using FTIR spectroscopy combined with chemometrics. Talanta Open. 2022;6:100168. doi: 10.1016/j.talo.2022.100168. [DOI] [Google Scholar]
  34. Vinciguerra L., Marcelo M. C. A., Motta T. M. C., Meneghini L. Z., Bergold A. M., Ferrão M. F.. Chemometric tools and FTIR-ATR spectroscopy applied in milk adulterated with cheese whey. New J. Chem. 2019;42(3):249–254. doi: 10.21577/0100-4042.20170337. [DOI] [Google Scholar]
  35. Kumar R., Sharma V.. Chemometrics in forensic science. Trends Anal. Chem. 2018;105:191–201. doi: 10.1016/j.trac.2018.05.010. [DOI] [Google Scholar]
  36. Soares S. F. C., Gomes A. A., Araujo M. C. U., Filho A. R. G., Galvão R. K. H.. The successive projections algorithm. Trends Anal. Chem. 2013;42:84–98. doi: 10.1016/j.trac.2012.09.006. [DOI] [Google Scholar]
  37. Yang B., Wang J., Zhang M., Shu H., Yu T., Zhang X., Yao W., Sun L.. A state-of-the-art survey of solid oxide fuel cell parameter identification: Modeling, methodology, and perspectives. Energy Convers. Manage. 2020;213:112856. doi: 10.1016/j.enconman.2020.112856. [DOI] [Google Scholar]
  38. Gomes A. A., Azcarate S. M., Diniz P. H. G. D., Fernandes D. D. S., Veras G.. Variable selection in the chemometric treatment of food data: A tutorial review. Food Chem. 2022;370:131072. doi: 10.1016/j.foodchem.2021.131072. [DOI] [PubMed] [Google Scholar]
  39. Bruneau A., de Queiroz L. P., Ringelberg J. J., Borges L. M., Bortoluzzi R. L. d. C., Brown G. K., Cardoso D. B. O. S., Clark R. P., Conceição A. d. S., Cota M. M. T.. et al. Advances in Legume Systematics 14. Classification of Caesalpinioideae. Part 2: Higher-level classification. Phytokeys. 2024;240:1–552. doi: 10.3897/phytokeys.240.101716. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. de Souza R. P., de Oliveira C. M. F., de Souza R. C. L., Leite L. L. L., Oliveira A. L. O., Ferreira J. V. B. P., de Freitas Pessoa A. M., Oliveira J. L. S., Meireles V. J. S., Lustosa Barros E. M., Viana Nunes A. M., Chevs M. H., do maral M. P. M., Vieira Júnior G. M.. Unraveling the metabolomic profile and acute toxicity of ethanolic extract from Mimosa tenuiflora (Willd.) Poir. root bark. Toxicon. 2024;249:108076. doi: 10.1016/j.toxicon.2024.108076. [DOI] [PubMed] [Google Scholar]
  41. do Nascimento I. A., Braz-Filho R., de Carvalho M. G., Mathias L., Fonseca F. d. A.. Flavonoids and other compounds isolated from Mimosa artemisiana Heringer and Paula. New J. Chem. 2012;35(11):2159–2164. doi: 10.1590/S0100-40422012001100010. [DOI] [Google Scholar]
  42. Danciu C., Pavel I. Z., Babuta R., Ersilia A., Oana S., Pop G., Soica C., Dehelean C., Radulov I.. Total phenolic content, FTIR analysis, and antiproliferative evaluation of lupine seeds harvested from western Romania. Ann. Agric. Environ. Med. 2017;24(4):726–731. doi: 10.26444/aaem/80795. [DOI] [PubMed] [Google Scholar]
  43. Jurasekova Z., Garcia-Ramos J. V., Domingo C., Sanchez-Cortes S.. Surface-enhanced Raman scattering of flavonoids. J. Raman Spectrosc. 2006;37:1239–1241. doi: 10.1002/jrs.1634. [DOI] [Google Scholar]
  44. Jurasekova Z., Domingo C., Garcia-Ramos J. V., Sanches-Cortez S.. Vibrational Spectroscopy as an Analytical Tool in the Identification and Characterization of Natural Dyes employed in the Cultural Heritage. Coalition. 2007;14:14–21. [Google Scholar]
  45. Segneasu, A. E. ; Gozescu, I. ; Dabici, A. ; Sfirloaga, P. ; Szabadai, Z. . Organic Compounds FT-IR Spectroscopy. In Macro to Nano Spectroscopy; Uddin, J. , Ed.; InTech, 2012. [Google Scholar]
  46. Rizwan K., Majeed I., Bilal M., Rasheed T., Shakeel A., Iqbal S.. Phytochemistry and Diverse Pharmacology of Genus Mimosa: A Review. Biomolecules. 2022;12:83. doi: 10.3390/biom12010083. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Okonkwo C. J., Njoku O. U., Okonkwo T. J. N., Afieroho O. E., Proksch P.. Two new acylated flavonol glycosides from Mimosa pigra L. leaves sub-family Mimosoideae. Future J. Pharm. Sci. 2016;2:71–75. doi: 10.1016/.j.fjps.2016.08.003. [DOI] [Google Scholar]
  48. Adurosakin O. E., Iweala E. J., Otike J. O., Dike E., Uche M. E., Owanta J. I., Ugbogu O. C., Chinedu S. N., Ugbogu E. A.. Ugbogu EA Ethnomedicinal uses, phytochemistry, pharmacological activities and toxicological effects of Mimosa pudica - A review. Pharmacol. Res.Mod. Chin. Med. 2023;7:100241. doi: 10.1016/j.prmcm.2023.100241. [DOI] [Google Scholar]

Articles from ACS Omega are provided here courtesy of American Chemical Society

RESOURCES