INTRODUCTION:

The genus Melastoma belonging to Melastomataceae family, consists of about 100 species of which 22 species have been well characterized in the Southeast Asian region. About ten species are present in Malaysia and Melastoma malabathricum or locally known in Malaysia as “Senduduk” or in China as “yemudan” or shanshiliu” is one of the most studied herbs among Melastoma species^1,2. It is a shrub with height about 1.5 to 5 m, and its flowers and fruits are quite characteristic of the genus. The plant is commonly found in roadsides, cleared areas or wasteland.

The plant has been reported to exhibit various medicinal properties in ethnobotanical^3,4 or bioactivity studies^5–9. Some of these activities have been related to phenolics such as flavonoids and tannins¹⁰ present in the plant and these compounds, as well as others, have been isolated or identified in the plant by simple phytochemical screening methods or more elaborate techniques of chromatography involving HPLC^11–13, GC^14,15 or TLC analyses¹⁶. Other than flavonoids the compounds belonged to major chemical groups like tannins¹⁷, amides¹⁸ and terpenoids. An important point with these techniques is whether they can serve as adequate quality control methods. These chemicals can serve as biomarkers for the plant but do they function as bioactive marker¹⁹, as a pharmacopoeia marker or both? A pharmacopeial marker of an herb may be ubiquitous²⁰ as for example, there have been more than 9000 known structures of flavones belonging to flavonoids or more than 10,000 alkaloids²¹. The important consideration is which of these groups or even particular compound in a group is to be chosen for quality control.

The quality of herbs is characterized by growing conditions related to locations, climate, as well as harvest time and plants with similar kinship or related families will tend to have similar properties and it is vital to differentiate among these varieties. The FTIR spectroscopy offers advantages compared to chromatographic techniques in terms of speed, simplicity, and safety. The FTIR, especially the attenuated total reflection type offers greater convenience and less sample preparation compared to the transmission analysis²².

This study aimed to use ATR-FTIR spectroscopy on authenticated Melastoma malabathricum samples from seven locations to establish a robust classification model for distinguishing the samples. The second derivative spectra were used, and three different pattern recognition methods or chemometrics^23,24 were used to select location-differentiating wavenumbers in the spectra. These variables were tested on different classification techniques to obtain a non-biased method that was efficient and low cost for geographical authentication of M. malabathricum.

MATERIALS AND METHODS:

This study consisted of three main approaches: sample preparation, FTIR analysis and data processing with chemometrics.

Chemicals

Samples of Melastoma plant materials

Melastoma malabathricum Linn samples consisting of leaf, flower, fruit, and stem were collected from natural habitats in two states of the northeast coast of Peninsula Malaysia between November 2013 and April 2014. The samples were coded as L1 to L7, and details of their sources are given in Table 1. The samples were identified by the morphological characteristics of the plant in reference to established monographs and references^25,26. The species of the samples were confirmed by a botanic scientific officer, Noor Haslinda Harun, and an herbarium sample with the specimen voucher 00245 was deposited in the herbarium unit, Faculty of Bioresources and Food Industry, Universiti Sultan Zainal Abidin. The collected leaves were washed thoroughly with distilled water to remove foreign matter and the samples were then oven-dried at 50℃ for 24 hours. The samples were finely powdered using an analytical grinder and sealed in a tight plastic bag at room temperature.

Table 1 The code and location of the Melastoma malabathricum samples

Code

Locality

State

Kuala Terengganu

(E:103.05430°, N5.40741°)

Kemaman (E:103.419249°, N4.228478°)

Jertih (E:102.51170°, N5.74639°)

Tumpat (E:102.15711°, N6.14106°)

Bakong Luar (E:102.08554°, N6.07746°)

Jeram Perdah (E:102.05292°, N6.07832°)

Bachok (E:102.39545°, N6.04336°)

Terengganu

Kelantan

ATR-FTIR spectral acquisition

The metabolite fingerprinting of M. malabatricum leaf samples from the different sources were carried out by FTIR spectral analysis. Five mg of powdered leaf sample was placed on the attenuated total reflectance (ATR) diamond and then pressed using the built-in pressure applicator of the IRPrestige 21 Fourier transform infrared spectrometer (Shimadzu corporation, Nagoya-ku, Kyoto, Japan). The instrument was operated in the region of 400-4000 cm^-1 and the scan technique was used at number of scans of 30 for each measurement and the infrared measurements were performed at the resolution of 4 cm^-1. A total of 20 replicate measurements were carried out for each location and the data were saved in ASCII format.

Data pre-processing

The spectral data obtained were used without smoothing but were derivatized (second derivative with number of points equalling 5) as it is an efficient method for eliminating baseline drifts and enhancing differences in the spectra ²⁷. Derivatization was carried out with IR Solution software Shimadzu version 1.40. The data obtained were then further processed by using tools in Microsoft® Excel 2000 (Microsoft Corp., WA, USA), the free R software and Matlab R2019b (The MathWorks Inc., Natick, MA, USA). The principal component analysis (PCA) and random forest (RF) were carried out using the prcomp and randomForest libraries in R, whereas LDA was carried out using MASS library in R. The genetic algorithm (GA) optimization technique was carried out using Matlab R2019b.

Chemometric techniques for wavenumber selection and classification of samples

The different chemometric techniques used to select the differentiating wavenumber for the FTIR-ATR data from various sources were PCA, RF and GA. The FTIR-ATR data used were in the region of 600 cm^-1to 2000 cm^-1 and these data consisted of a 120 × 871 data matrix, where each row represented the sample and each column the spectral data at a given wavenumber. PCA was implemented using the function ‘prcomp’ in the R package and variables with the highest loadings for the first principal component were selected. The RF method was carried out with the function randomForest and as this method can function as a classification technique simultaneously, the data was partitioned into a training set and testing set in a ratio of 2:1. The selection of variables was done on the training set, and it needed two parameters to be set. These were ntree, which was set to 500 and mtry, which was set to 30. The mean decrease in accuracy was used to select the wavenumber variables for classification of location. The GA was carried out using Matlab and the approach here was slightly modified from that of the work with Phyllanthus samples ^28,29. In this work, the FTIR data were divided into Xdata and Xtune. The Xdata used the fitness function to find the wavenumber variables that maximize the ratio of between-groups variance to within-groups variance as in canonical variate analysis. However, the modification from the previous papers is that this ratio was calculated using a separate calibration data designated Xtune which avoids overfitting³⁰. The parameters for GA were numbers of chromosomes of 300, numbers of generations of 50 and numbers of canonical variate loadings of 4 were calculated. The twelve wavenumber variables that were chosen by GA for the classification of geographical location were from the first canonical variate loadings.

The accuracies of the twelve parameters chosen by each of the three selection algorithms were evaluated on the test set. The correct classifications were predicted using four different supervised chemometric techniques: k-nearest neighbors (kNN), linear discriminant analysis, Naïve-Bayes, and support vector machines (SVM). The kNN used numbers of neighbours of five with Euclidean distance for considering the correct classification. The other functions such as SVM used radial basis function instead of linear and the two parameters needed for it used default values. These were the cost function that was set to 1 and gamma used was 0.0833 (the inverse of the number of parameters). The classification performance for each location as given by the best discriminatory technique was estimated by three terms which were sensitivity, specificity, and efficiency³¹. These parameters were calculated from four values which are true positive (TP), true negative (TN), false positive (FP) and false negative (FN). Positive class also means itself and negative class means other classes and the earlier four terms are defined as following:

TP: correctly identified samples of positive class (or itself).

TN: correctly identified samples of negative class (or other class).

FP: incorrectly identified samples of negative class.

FN: incorrectly identified samples of negative class.

Sensitivity=TP/(TP + FN)

Specificity = TN/(TN + FP)

Efficiency =√SRNS x SPEC

RESULTS:

Plant morphology

The variety of Melastoma malabathricum used in this study was identified by the appearance of leaves, flowers and fruits and the samples of these are shown in Figure 1. The flowers had dark purple-magenta petals and the leaves were measuring between 12.5 - 16 cm in length and 2.9 - 4.5 in width.

Figure 1 Sample of leaf, fruit, and flower of Melastoma malabathricum.

FTIR-ATR spectra

The ATR-FTIR spectra of the Melastoma malabathricum for each representative sample from the seven locations are shown in Figure 2. The spectra showed that the seven locations have similar transmittance patterns, and the numbers of peaks are about the same and they differ in intensity and shape of the peaks. There are broad bands at 3200-3440 cm^-1 due to stretching of the hydroxyl group from carbohydrates. The bands located between 2840 cm^-1 and 2950 cm^-1 might be associated with the C–H stretching of alkanes and alkenes. The fingerprint region of 850-1850 cm^-1 also shows few distinctive peaks. The prominent peak in all sample between 1040 and 1200 cm^-1 is suggested to belong to the stretching vibration of C-O in carbohydrate³².

The amplification of small differences in the spectra and enhancement of spectral resolution is obtained using second-derivative spectra which are shown in Figure 3. The major stable and visible peaks in the spectra are ten reverse peaks in the fingerprint region of 1000-1800 cm^-1 as well as the two peaks at around 2848 and 2916 cm^-1. The peaks at 1458 and 1475 cm^-1 are due to the skeletal vibration of aromatic hydrocarbons involving stretching with the ring.

Figure 2 ATR-FTIR spectra of seven representative locations.

The unsupervised pattern recognition technique of PCA was used to reduce the dimensionality of the data and obtain clustering among the samples. The clustering of the samples is shown in a two-dimensional scatter plot in Figure 4. The first principal component represented 20.4% of the total variance. In comparison, the second principal component represented 13.0% of the total variance while a total of 19 principal components were required to reach the total cumulative percentage of variance above 90%. The wavenumber variables chosen for classification of the samples were 1458, 1456, 1653, 1684, 1379, 1663, 1670, 1668, 1464, 1661, 1682, 1558 cm^-1 and these were the ones with the highest loadings or contribution for the first principal component.

Figure 3 The second derivative ATR-FTIR spectra of seven representative locations in region 400-3000 cm^-1showing some prominent peaks.

The selection of variables of the random forest was made robust by partitioning the data into a training set and tuning set with the former being used to select the variables. The random forest classifier was built using the out-of-bag (OOB) or leave-one-out method and the OOB estimation of error rate was 6.59%. The wavenumber variables selected by the random forest classifier based on the mean decrease in accuracy were 1175, 1373, 1173, 1053, 1371, 1049, 1055, 1369, 1344, 802, 1051 and 1356 cm^-1. The wavenumber variables with the highest magnitude in loadings in the first canonical variate of GA were selected for differentiating samples according to locations and these were 972, 756, 822, 735, 1497, 1315, 1200, 1721, 1342, 1045, 889, and 922 cm^-1.

The twelve variables selected by PCA, RF and GA were assessed for their ability to discriminate the origin of the samples and four different classifiers were used to avoid biasness related to a particular technique. The results of the correct classification using kNN, NB, SVM and LDA for the test set for the three sets of twelve variables are shown in Table 2 and the variables chosen by RF had the highest classification in all the respective algorithms. The best discriminatory algorithm among the four was LDA. The percentages of explained variance for each of the first linear discriminant for the variables chosen by RF, GA and PCA were 54.06, 40.79 and 53.48%.

The classification performance of sensitivity, specificity, and efficiency for the different locations based on the LDA as the best discriminatory technique for the variables selected by PCA, RF and GA are showed in Table 3. Four locations had 100% values for both sensitivity and specificity with LDA using variables selected by RF. The use of the parameter efficiency that combines the ability to distinguish samples belonging to that class and the capacity to exclude samples belonging to other class seems a better approach to compare the chemometric methods. The RF had the highest values for efficiency among the three methods for all locations other than L5, and this location had the lowest total value of the three selection methods, which showed it was the hardest to classify.

Table 2 Percentage of correct classifications for test samples by different classifiers for variables selected by PCA, RF and GA

Classification technique	Percentage of correct classification on the test set by
	PCA	RF	GA
kNN	53.1	75.5	73.5
NB	57.1	83.7	73.5
SVM	65.3	91.8	81.6
LDA	79.6	93.9	89.8

Table 3 The classification parameters by LDA for each location based on variables selected by three chemometric techniques.

Method	Parameter	Location
Method	Parameter	1	2	3	4	5	6	7
PCA	Sensitivity	85.7	71.4	71.4	85.7	57.1	100	85.7
	Specificity	95.2	100	95.2	88.1	100	95.2	100
	Efficiency	90.3	84.5	82.4	86.9	75.6	97.6	92.6
RF	Sensitivity	100	85.7	100	100	71.4	100	100
	Specificity	100	100	100	100	97.6	95.2	100
	Efficiency	100	92.6	100	100	84.5	97.6	100
GA	Sensitivity	85.7	85.7	100	100	85.7	71.4	100
	Specificity	100	100	97.6	97.6	95.2	97.6	100
	Efficiency	92.6	92.6	98.8	98.8	90.3	83.5	100

DISCUSSION:

One of the most prominent challenges in the herbal industry is the quality control of the sample and the origin of a sample plays an important part in the quality of an herb. The M. malabathricum used in this study is considered the synonym with M. affine, M. denticulatum and M. polyanthum^33,34. The sampling of the plant also considered the morphological differences with M. sanguineum, M. crinitum and M. imbricatum^1,35, and such knowledge or foresight would ensure that the correct samples were obtained. This study shows that ATR FTIR spectroscopy of the authenticated plant when combined with chemometric analysis, enables identification of the sample origin. The use of second derivative spectra had the advantage of resolving overlapping bands as well as gives a narrow band^36,37, and, when used with chemometric enables identification of the wavelengths responsible for differentiating samples. Three chemometric techniques were used for identification of the differentiating wavenumber. The first was PCA which is an unsupervised learning method requiring no prior knowledge of class structure. In contrast, the RF technique used enabled variable selection as well as the classification of samples. Among the variables selected by the three approaches, it is interesting to note that PCA selected a band at 1379, whereas RF selected two bands at 1371 and 1373, and the three of these are very close to 1375. The band at 1375 is one of nine bands in Gentiana rigescens that occurs in flower, leaf, stem, and root of the plant³⁸. Bands around 1375 could be due to C─H stretch deformation in hemicellulose, or cellulose or C─N stretch in aromatic amines³⁹. These could be why the methods selected few wavenumbers in that region since they were necessary to distinguishing the samples. Although GA did not have any wavenumber between 1360-1380 selected in the top fifty variables with the highest magnitude of loadings, the band 1375 cm^-1 was ranked 64^thor within the top 10% of the whole variables. The selected variables from these methods were used for the prediction of origin of samples using different classification techniques. The LDA had the highest correct classification for the test samples. The use of kNN for classification gave the least correct classification. This differed from some studies such as by Li and coauthors²⁷ on Marsdenia tenacissima where kNN performed as well as SVM for classification with all test samples being correctly classified. The use of grid search to optimize parameters for SVM could be done if this study used a higher number of samples or if LDA had performed as an inferior technique with any of the three sets of twelve variables chosen. The best variable selection method was RF. These results are not surprising as it is a very versatile technique that avoids overfitting as it considers only a small number of variables (the square root of the number of variables)⁴⁰ at a time out of a large number of predictors. It also works well with non-linear data. The differentiating variable for RF can be obtained directly by using the mean decrease in accuracy or gini and often these return similar variables. The variables from PCA were the ones with the highest loadings in the first principal component and although this principal component contributed more than the second, it might not be fully optimized for the purpose of classification. The variables from GA performed better as they were obtained from the first canonical variate loading that had a role in differentiating the sample according to origins. The use of these variables for distinguishing the samples showed that L5 was the hardest to differentiate whereas L7 was the easiest. The sample from L7 was found to contain a higher yield and this might be related to its diverse chemical content⁴¹ which could have played a role.

CONCLUSION

The species authentication of herbs and its geographical origin discrimination are vital aspects of quality control of herbal medicine. This paper examined the use of second derivative spectra of ATR-FTIR spectroscopy of morphologically authenticated Melastoma malabatricum samples to distinguish their properties among seven locations. Three different chemometric techniques were used to select twelve wavenumbers for further classification. RF gave the best results where the variables selected by it correctly classified 93.9% of the test samples using LDA. RF selected among its variables two wavenumbers in region 1371-1379 cm^-1 which is suggested to be related to cellulose or aromatic amine compounds in plants. The RF method had 100% specificity and sensitivity for four locations, of which one had the highest yield. The use of different chemometric selection method and running different classification techniques avoid bias and this method which is cheap and environmentally friendly could be used for other species and locations.

FUNDING:

The study was supported by Research Acculturation Collaborative Effort (RACE) Grant Scheme (no.RACE/F2/SKK/UniSZA/1), Ministry of Higher Education of Malaysia.

DECLARATION OF COMPETING INTEREST:

The authors declare that there is no conflict of interest regarding the publication of this article.

CREDIT AUTHORSHIP CONTRIBUTION STATEMENT:

AWA collected the samples, carried out the experiments, and assisted the data analysis, and wrote part of the manuscript. KSM was involved in sample identification, and herbal quality aspect together with SS and MRUS. SD conceived the study, checked the data analysis and wrote part of the manuscript. All authors read and approved the final manuscript.

AVAILABILITY OF DATA AND MATERIALS:

The processed data and algorithms for the multivariate analyses can also be obtained from corresponding author on reasonable request.

ACKNOWLEDGMENT:

The authors would like to express sincere gratitude to Marwan Saad Abdulrahman Azzubaidi from Faculty of Medicine, UniSZA for final proofreading.

REFERENCES:

1. Rajenderan MT. Ethno medicinal uses and antimicrobial properties of Melastoma malabathricum. SEGI Rev. 2010;3(2):34-44.

2. Zheng WJ, Ren YS, Wu ML, et al. A review of the traditional uses, phytochemistry and biological activities of the Melastoma genus. J Ethnopharmacol. 2021; 264(July 2020): 113322. doi:10.1016/j.jep.2020.113322

3. Ong HC, Zuki RM, Milow P. Traditional knowledge of medicinal plants among the Malay villagers in Kampung Mak Kemas, Terengganu, Malaysia. Ethno-Medicine. 2011; 5(3): 175-185.

4. Neamsuvan O, Sengnon N, Seemaphrik N, Chouychoo M, Rungrat R, Bunrasri S. A survey of medicinal plants around upper Songkhla Lake, Thailand. African J Tradit Complement Altern Med. 2015; 12(2): 133-143.

5. Sulaiman MR, Somchit MN, Israf DA, Ahmad Z, Moin S. Antinociceptive effect of Melastoma malabathricum ethanolic extract in mice. Fitoterapia. 2004; 75(7-8): 667-672. doi:10.1016/j.fitote.2004.07.002

6. Balamurugan K, Sakthidevi G, Mohan VR. Anti-inflammatory activity of leaf of Melastoma malabathricum L. (Melastomataceae). Int J Res Ayurveda Pharm. 2012; 3(6): 801-802. doi:10.7897/2277-4343.03622

7. Nurdiana S, Marziana N. Wound healing activities of Melastoma malabathricum leaves extract in Sprague Dawley rats. Int J Pharm Sci Rev Res. 2013; 20(2): 20-23.

8. Balamurugan K, Nishanthini A, Mohan VR. Antidiabetic and antihyperlipidaemic activity of ethanol extract Melastoma malabathricum Linn. leaf in alloxan induced diabetic rats. Asian Pac J Trop Med. 2014; 4(Suppl 1): S442-S448. doi:10.12980/APJTB.4.2014C122

9. Kumar V, Sachan R, Rahman M, et al. Chemopreventive effects of Melastoma malabathricum L. extract in mammary tumor model via inhibition of oxidative stress and inflammatory cytokines. Biomed Pharmacother. 2021; 137: 111298. doi:10.1016/j.biopha.2021.111298

10. Hamid HA, Ramli ANM, Zamri N, Yusoff MM. UPLC-QTOF/MS-based phenolic profiling of Melastomaceae, their antioxidant activity and cytotoxic effects against human breast cancer cell MDA-MB-231. Food Chem. 2018; 265(February): 253-259. doi:10.1016/j.foodchem.2018.05.033

11. Mamat SS, Kamarolzaman MFF, Yahya F, et al. Methanol extract of Melastoma malabathricum leaves exerted antioxidant and liver protective activity in rats. BMC Complement Altern Med. 2013;13:326. doi:10.1186/1472-6882-13-326

12. Karupiah S, Ismail Z. Antioxidative effect of Melastoma malabathticum L extract and determination of its bioactive flavonoids from various location in Malaysia by RP-HPLC with diode array detection. J Appl Pharm Sci. 2013; 3(2): 19-24. doi:10.7324/JAPS.2013.30204

13. Zakaria ZA, Raden Mohd Nor RNS, Hanan Kumar G, et al. Antinociceptive, anti-inflammatory and antipyretic properties of Melastoma malabathricum leaves aqueous extract in experimental animals. Can J Physiol Pharmacol. 2006;84(12):1291-1299. doi:10.1139/Y06-083

14. Balamurugan K, Nishanthini A, Lalitharani S, Mohan VR. GC-MS Determination of bioactive components of Melastoma malabathricum L. Int J Curr Pharm Res. 2012; 4(4): 24-26.

15. Zakaria ZA, Jaios ES, Omar MH, et al. Antinociception of petroleum ether fraction derived from crude methanol extract of Melastoma malabathricum leaves and its possible mechanisms of action in animal models. BMC Complement Altern Med. 2016; 16(1): 488. doi:10.1186/s12906-016-1478-1

16. Kader MA, Rahman MM, Mahmud S, Khan MS, Mukta S, Zohora FT. A comparative study on the Antihyperlipidemic and antibacterial potency of the shoot and flower extracts of Melastoma malabathricum Linn’s. Clin Phytoscience. 2023; 9(1): 5. doi:10.1186/s40816-023-00355-6

17. Yoshida T, Nakata F, Hosotani K, Nitta A, Okuda T. Tannins and related polyphenols of Melastomataceous plants. V. Three new complex tannins from Melastoma malabathricum L. Chem Pharm Bull. 1992; 40(7): 1727-1732.

18. Susanti D, Sirat HM, Ahmad F, Ali RM. Bioactive constituents from the leaves of Melastoma malabathricum L. J Ilm Farm. 2008; 5(1): 1-8.

19. Lau CBS, Yue GGL, Lau KM, et al. Method establishment for upgrading chemical markers in pharmacopoeia to bioactive markers for biological standardization of traditional Chinese medicine. J Tradit Complement Med. 2019; 9(3): 179-183. doi:10.1016/j.jtcme.2018.09.003

20. Houriet J, Allard PM, Queiroz EF, et al. A mass spectrometry based metabolite profiling workflow for selecting abundant specific markers and their structurally related multi-component signatures in Traditional Chinese Medicine multi‐herb formulae. Front Pharmacol. 2020; 11(December): 1-23. doi:10.3389/fphar.2020.578346

21. Pan SY, Zhou SF, Gao SH, et al. New perspectives on how to discover drugs from herbal medicines: CAM’S outstanding contribution to modern therapeutics. Evidence-based Complement Altern Med. 2013; 2013. doi:10.1155/2013/627375

22. Wang Y, Huang HY, Zuo ZT, Wang YZ. Comprehensive quality assessment of Dendrubium officinale using ATR-FTIR spectroscopy combined with random forest and support vector machine regression. Spectrochim Acta - Part A Mol Biomol Spectrosc. 2018; 205: 637-648. doi:10.1016/j.saa.2018.07.086

23. Shiyan S, Ramadona N, Utami WF, Depriyanti N, Mukafi A, Noviandhani W. Preparation and FTIR-ATR combined with chemometrics analysis of self-emulsifying loaded sungkai extract from Peronema canecens. Res J Pharm Technol. 2023; 16(1): 79-85. doi:10.52711/0974-360X.2023.00014

24. Fatmarahmi DC, Susidarti RA, Swasono RT, Rohman A. Identification and quantification of metamizole in traditional herbal medicines using spectroscopy ftir-atr combined with chemometrics. Res J Pharm Technol. 2021; 14(8): 4413-4419. doi:10.52711/0974-360X.2021.00766

25. van Valkenberg JLCH, Bunyapraphatsara N. Melastoma malabathricum L. In: van Valkenberg JLCH, Bunyapraphatsara N, eds. Plant Resources of South-East Asia. 2001; 12(2).

26. Joffry SM, Yob NJ, Rofiee MS, et al. Melastoma malabathricum (L.) Smith ethnomedicinal uses, chemical constituents, and pharmacological properties: A review. Evidence-based Complement Altern Med. 2012; 2012. doi:10.1155/2012/258434

27. Li C, Yang SC, Guo QS, Zheng KY, Wang PL, Meng ZG. Geographical traceability of Marsdenia tenacissima by Fourier transform infrared spectroscopy and chemometrics. Spectrochim Acta - Part A Mol Biomol Spectrosc. 2016; 152: 391-396. doi:10.1016/j.saa.2015.07.086

28. Dharmaraj S, Gam LY, Sulaiman SF, Mansor SM, Ismail Z. The application of pattern recognition techniques in metabolite fingerprinting of six different Phyllanthus spp. Spectroscopy. 2011; 26(1): 69-78. doi:10.3233/SPE-2011-0527

29. Dharmaraj S, Jamaludin AS, Razak HM, et al. The classification of Phyllanthus niruri Linn. according to location by infrared spectroscopy. Vib Spectrosc. 2006; 41(1): 68-72.

30. Kemsley EK. A genetic algorithm (GA) approach to the calculation of canonical variates (CVs). Trends Anal Chem. 1998; 17(1): 24-34.

31. Li Y, Zhang JY, Wang YZ. FT-MIR and NIR spectral data fusion: a synergetic strategy for the geographical traceability of Panax notoginseng. Anal Bioanal Chem. Published online. 2017: 1-13. doi:10.1007/s00216-017-0692-0

32. Joshi DD. FTIR Spectroscopy. In: Herbal Drugs and Fingerprints: Evidence Based Herbal Drugs. Springer India; 2012: 121-146. doi:10.1007/978-81-322-0804-4

33. GBIF Secretariat. Melastoma malabathricum L. Global Information Hub on Integrated Medicine. Published 2014. https://www.globinmed.com/medicinal_herbs/melastoma-malabathricum-l-104890/

34. POWO. Melastoma malabathricum L. Plants of the World Online. Facilitated by the Royal Botanic Gardens, Kew. Published 2024. Accessed April 8, 2024. http://www.plantsoftheworldonline.org/

35. Meyer K. Revision of the Southeast Asian genus Melastoma (Melastomataceae). Blumea. 2001; 46(2): 351-398.

36. Rohaeti E, Rafi M, Syafitri UD, Heryanto R. Fourier transform infrared spectroscopy combined with chemometrics for discrimination of Curcuma longa, Curcuma xanthorrhiza and Zingiber cassumunar. Spectrochim Acta - Part A Mol Biomol Spectrosc. 2015; 137: 1244-1249. doi:10.1016/j.saa.2014.08.139

37. Wang YY, Li JQ, Liu HG, Wang YZ. Attenuated total reflection-Fourier transform infrared spectroscopy (ATR-FTIR) combined with chemometrics methods for the classification of Lingzhi species. Molecules. 2019; 24(12): 2210. doi:10.3390/molecules24122210

38. Wang Y, Shen T, Zhang J, Huang HY, Wang YZ. Geographical authentication of Gentiana rigescens by high-performance liquid chromatography and infrared spectroscopy. Anal Lett. 2018;51(14):2173-2191. doi:10.1080/00032719.2017.1416622

39. Brangule A, Šukele R, Bandere D. Herbal medicine characterization perspectives using advanced FTIR sample techniques – diffuse reflectance (DRIFT) and photoacoustic spectroscopy (PAS). Front Plant Sci. 2020; 11(April): 356. doi:10.3389/fpls.2020.00356

40. de Santana FB, Mazivila SJ, Gontijo LC, Neto WB, Poppi RJ. Rapid discrimination between authentic and adulterated Andiroba oil Using FTIR-HATR spectroscopy and random forest. Food Anal Methods. 2018; 11(7): 1927-1935. doi:10.1007/s12161-017-1142-5

41. Wan-Azemin A. Investigation of chemical profile and apoptosis induction by Melastoma malabathricum L . extracts against HEPG2 cell lines. Universiti Sultan Zainal Abidin; 2017.

Received on 20.05.2024 Modified on 30.06.2024

Research J. Pharm. and Tech 2024; 17(8):3769-3776.

DOI: 10.52711/0974-360X.2024.00586