Get 20M+ Full-Text Papers For Less Than $1.50/day. Subscribe now for You or Your Team.

Learn More →

Data Mining Methods for Omics and Knowledge of Crude Medicinal Plants toward Big Data Biology

Data Mining Methods for Omics and Knowledge of Crude Medicinal Plants toward Big Data Biology Volume No: 4, Issue: 5, January 2013, e201301010, http://dx.doi.org/10.5936/csbj.201301010 CSBJ Data Mining Methods for Omics and Knowledge of Crude Medicinal Plants toward Big Data Biology a,b a a d c a a Farit M. Afendi , Naoaki Ono , Yukiko Nakamura , Kensuke Nakamura , Latifah K. Darusman , Nelson Kibinge , Aki Hirai Morita , Ken e f a a,* Tanaka , Hisayuki Horai , Md. Altaf-Ul-Amin , Shigehiko Kanaya Abstract: Molecular biological data has rapidly increased with the recent progress of the Omics fields, e.g., genomics, transcriptomics, proteomics and metabolomics that necessitates the development of databases and methods for efficient storage, retrieval, integration and analysis of massive data. The present study reviews the usage of KNApSAcK Family DB in metabolomics and related area, discusses several statistical methods for handling multivariate data and shows their application on Indonesian blended herbal medicines (Jamu) as a case study. Exploration using Biplot reveals many plants are rarely utilized while some plants are highly utilized toward specific efficacy. Furthermore, the ingredients of Jamu formulas are modeled using Partial Least Squares Discriminant Analysis (PLS-DA) in order to predict their efficacy. The plants used in each Jamu medicine served as the predictors, whereas the efficacy of each Jamu provided the responses. This model produces 71.6% correct classification in predicting efficacy. Permutation test then is used to determine plants that serve as main ingredients in Jamu formula by evaluating the significance of the PLS-DA coefficients. Next, in order to explain the role of plants that serve as main ingredients in Jamu medicines, information of pharmacological activity of the plants is added to the predictor block. Then N-PLS-DA model, multiway version of PLS-DA, is utilized to handle the three-dimensional array of the predictor block. The resulting N-PLS-DA model reveals that the effects of some pharmacological activities are specific for certain efficacy and the other activities are diverse toward many efficacies. Mathematical modeling introduced in the present study can be utilized in global analysis of big data targeting to reveal the underlying biology. 1. Introduction Data-intensive sciences have progressed in modern astronomy [1], the rapid increasing of omics data produced by genomics, biology [2-8], computational materials science [9], ecology [10-11] transcriptomics, proteomics and metabolomics [2-8]. This situation and social science [12] because open-access data has increased is also a feature of the ethnomedicinal survey and the number of drastically. Data-intensive or -driven discovery in biology requires a medicinal plants is estimated to be 40,000 to 70,000 around the large open pool of data across the full breadth of the life sciences and world [13] and many countries utilize these plants as blended herbal the access to the pool will invite “New” logic, strategies and tools to medicines, e.g., China (traditional Chinese medicine), Japan (Kampo discover new trends, associations, discontinuities, and exceptions that medicine), India (Ayruveda, Siddha and Unani) and Indonesia reveal aspects of the underlying biology [2, 5, 6]. Big data biology, (Jamu). Blended herbal medicines as well as single herb medicines which is a discipline of data-intensive science, was proposed based on include a large number of constituent substances which exert effects on human physiology through a variety of biological pathways. To comprehensively understand the medicinal usage of plants based upon traditional and modern knowledge, we add to KNApSAcK Family database systems the selected herbal ingredients i.e., the formulas of Graduate School of Information Science, Nara Institute of Science and Kampo and Jamu, omics information in plants and humans, and Technology, Nara 630-0101, Ikoma, Japan b physiological activities in humans [14-16]. These information need to Department of Statistics, Bogor Agricultural University, Jln. Meranti, be connected in a way that enables scientists to make predictions Kampus IPB Darmaga, Bogor 16680, Indonesia based on general principles. Biopharmaca Research Center, Bogor Agricultural University, Kampas IPB In this mini-review, we discuss the usage of KNApSAcK Family Taman Kencana, Jln. Taman Kencana No. 3 Bogor 16151, Indonesia DB in metabolomics, explain mining techniques such as principal Maebashi Institute of technology, 450-1 Kamisadori, Maebashi-shi, component analysis (PCA), partial least square regression (PLSR) and Gunma, 371-0816 Japan multiway model, and show their application on Indonesian blended Department of Medicinal Resources, Institute of Natural Medicine, herbal medicines (Jamu) as a case study. University of Toyama, 2630 Toyama, 930-0194, Japan Department of Electronic and Computer Engineering, Ibaraki National 2. KNApSAcK Family Database College of Technology, 866 Nakane, Hitachinaka, Ibaraki 312-8508, Japan Omics biology, like most scientific disciplines, is in an era of * Corresponding author. accelerated increase of data, so called big data biology [2-8]. Large- E-mail address: skanaya@gtc.naist.jp (Shigehiko Kanaya) scale sequencing centers, high-throughput analytical facilities and Data Mining Methods for Omics individual laboratories produce vast amounts of data such as planning to translate them into English as early as possible. Lunch nucleotide and protein sequences, gene expression measurements, Box DB comprises information on 800 edible species which include protein and genetic interactions, mass spectra of metabolites and the species introduced to Japan from outside or originally grown in phenotype studies. The goal of investigating the interactions between Japan, general information of the crops and the effect of them on medicinal/edible plants and humans is to comprehensively understand human health. the molecular mechanism of medicinal plants on human physiology Noncommunicable diseases such as heart disease, metabolic based on current and traditional knowledge. Optimization of blended disease, cancer and respiratory disease, which superseded the herbal formulas should be developing using information derived from infectious diseases because of the development and widespread plant and human omics. To reach this goal we need to develop distribution of vaccines and antimicrobial drugs, account for 60% of databases based on the platform shown in Fig. 1A. KNApSAcK all deaths worldwide and 80% of deaths in low- and middle-income family DBs have been developed for this purpose [14-16]. Relations countries [17]. Food and ingredients in sanative diet and more among individual DBs are illustrated in Fig. 1A and main page of effective combination of foods beneficial against those KNApSAcK Family DB is shown in Fig. 1B. noncommunicable diseases are accumulated in DietNavi and DietDish DBs, respectively (b and d in Fig. 1). FoodProcessor DB comprises 309 retortable pouch foods encompassed by 261 food ingredients produced in Japan, and connected with DietNavi and KNApSAcK core by species names of foods. To systematize crude drugs by multifaceted view points, we have developed four DBs (WorldMap, KAMPO, JAMU and TeaPot DBs as shown in e-h of Fig. 1). The KNApSAcK WorldMap DB comprises 46,256 geographic zone-plant pair entries in 217 geographical zones except mini-states such as the Principalities of Liechtenstein, Monaco and Andorra, and the Vatican City. Prescriptions corresponding to Japanese and Indonesian herbal medicines have been accumulated in KAMPO and JAMU DBs, respectively. KAMPO DB is comprised of 1,581 primary formulas classified in to 336 formula names encompassed by 278 medicinal plants which are approved by the National health insurance authority in Japan. JAMU DB is comprised of 5,310 formulas encompassed by 550 medicinal plants and 12 anatomical regions which are approved by the National Agency of Drug and Food Control (NA-DFC) of Indonesia. Medicinal/edible plants reported in the scientific literature have been classified into geographic zones using the International Organization for Standardization (ISO3166), which defines geographic zones based on the borders between nations and small islands. Herbs are defined as any plants with leaves, seeds, and flowers used for flavoring, food, medicine, perfume and parts of such a plant as used in cooking. Those are accumulated in TeaPot DB. Two types of biological activities, that is, activities of natural resources and metabolites to other species including human, i.e., antibiotic, anticancer and so on are accumulated in Natural Activity and Metabolite Activity DBs (Fig. 1B), respectively. The former and the latter comprised 33,703 and 6,677 entries, respectively. For extension of species-metabolite relationship DB to metabolic pathways, it is needed to design secondary metabolic pathway DB for detection of metabolic pathways based on enzyme reactions and prediction of reactions by peptide sequences. So we have developed Motoercycle DB containing 2,421 entries. The metabolomics of plants is developing rapidly [18-20 and references in Table 1], and it will be an important topic in the systems-biological studies of interactions between plants and humans, which is included in the topics of big data biology [2-8], with the goal of achieving a holistic understanding of plant function and healthcare, including the activity Figure 1. Integrated platform of knowledge of medicinal plants and plant of medicinal plants as well as interaction between plants and their and human –omics and KNApSacK Family databases. (A) The relations of environment [14-16, 21, 22]. attributes among individual DBs. (B) Main window of KNApSAcK Family To facilitate access to metabolite information obtained from DB, indexes from a to i in panel A correspond to those in panel B. analytical techniques, we have developed species-metabolite relationship DB (KNApSAcK Core DB) which contains 106,418 species-metabolite relationships encompassing 21,705 species and Four DBs (Lunch Box DB, DietNavi DB, Food Processor DB and 50,897 metabolites. Nine databases of KNApSAcK family (except DietDish DB, a-d in Fig. 1) are about Food & Health related with DietDish) are connected with KNApSAcK Core DB to easily obtain Japanese foods and ingredients explained in Japanese language because candidates of secondary metabolites in species utilized in several initially we developed them targeting the Japanese people, but we are purposes [23]. The KNApSAcK Core DB was utilized in very Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics diverged purposes of metabolomics studies including identification of Consider a data matrix A = (a1 a2 … ap) with n observations and metabolites (‘Exp’ in Table 1), construction of integrated databases let V (p x p) be the variance-covariance matrix of A. The principal (‘DB’), bioinformatics and systems biology (‘Bioinfo’) , and cited in at components of A, Z = (z1 z2 … zp), are calculated as least 110 papers listed in Table 1, that is, in 29 papers in the period of 2006-2008, 25 papers in the period of 2009, 20 papers in 2010, z = Ac ( j = 1, 2, …, p) j j 18 papers in 2011, 18 papers in 2012-2013. In addition, it was applied in diverged species from bacteria to plants and animals, in where cj is the j-th eigenvector of V which correspond to the j-th total 28 species, that is, Angelica acutiloba [74], Arabidopsis lyrata eigenvalue of V (j). The properties of PCs are: (1) Var(zj) = j; (2) ssp. petraea [56], Arabidopsis thaliana[25, 30, 33, 35, 37, 46, 47, 62, Cov(zj,zj’) = 0, j  j’; (3) Var(z1)  Var(z2)  … Var(zp). The 70, 86, 99, 103, 104, 108, 109, 121, 122], Atriplex halimus [127], cumulative proportion of variance of the original variables explained Bacillus subtilis [113], Brassica oleraceae var capitata [60], Brufelsia by the first J principal components can be obtained as calycina [81], Capsicum sp. [123], Citrus sinensis [131], Curcuma longa [77], Ephedra sp. [67], Escherichia coli [51], Fragaria x ananassa [40, 43, 44], Fragaria vesca [105], Glycine max [53],   j j1 Glycyrrhiza uralensis [94], Hordeum vulgare [80, 102], Homo Pr(z ) sapiens [63, 101], Jatropha curcas [124, 125], Malx x domestica  j [126], Ophiorrhiza pumila [117], Oryza sativa [49, 61], Papaver j1 somniferum [42], Rattus norvegicus [39, 97], Rizotania solani [79], Solanum lycopersicum [45, 48], Solanum tuberosum [98] and Zea mays [120]. PLSR is a regression method, which assumes underlying factors In the period of 2006-2008, many review papers [‘Review’ in among the predictors account for most of the response variation [133, Table 1] focused on metabolomics platforms integrated by mass- 134]. These underlying factors of X-variate spectrometry and metabolite databases including KNApSAcK Core [29, 31, 34, 38, 42, 49, 52] and on linking chemistry with biology T = XW [24], and on metabolome researches targeting the model plant Arabidopsis thaliana [30, 33, 35, 37]. In 2009, metabolome studies are obtained by maximizing their covariance with the corresponding were extended to diverged species such as crops and medicinal plants underlying factors of Y-variate where X is an n  m matrix of [53, 60, 61, 67, 68, 73, 74, 78] and to engineering studies such as quality assessment based on metabolomics [73, 74]. Thus predictors, Y is an n  p matrix of responses, T is an n  c matrix of metabolomics was applied from model species to crops and medicinal X-score factors, and W is m  c matrix of weight. Note that n is the herbs. In the period of 2010-2013, metabolomics was further number of observations, m is the number of predictors, p is the extended to genetics such as QTL [80, 98, 126], and to explanation number of responses, and c is the number of components. of species by metabolites, i.e., ecological subjects [85] phytoalexins The X-score factors, i.e. matrix T, have the following [119], herbivore-induced metabolites [120] and defense against properties [133]. pathogens [131], and to stress responses [115, 116, 127]. In addition, metabolomics has also been tried in imaging studies [112, 129]. a. When multiplied by loadings P, they are good summaries of X, Species-metabolite relation database KNApSAcK Core has been i.e. the X-residuals E are small utilized in the extended fields of metabolomics researches and the horizon of metabolomics researches could be recognized by reviewing X = TP + E the works that utilized and/or cited the KNApSAcK DB. Methodologies for multivariate analysis to statistically process the b. The X-score factors are good predictors of Y, i.e. massive amount of metabolome data were reviewed in [16] and to systematize blended herbal medicines in Kampo [15]. In the following Y = TQ + F section, we focus on the mining studies of blended herbal medicines for systematically understanding the composition of medicinal herbs The Y-residuals F express the deviations between the observed and to efficacies on humans, that is, principal component analysis (PCA) modeled responses. that makes it possible to systematize the ingredient in individual Based on Eq. (3), Eq. (5) can be rewritten as a multiple regression blending systems, partial least squares (PLS) that can relate the model ingredients of medicinal herbs to the efficacies and N-PLS that can connect multi-factors to the efficacies. We initially explain individual Y = XWQ + F = XB + F techniques in Section 3 and then discuss their application in data- mining of blended types of herbal medicines in Section 4. Thus, PLSR coefficients B can be written as 3. Mathematical Methods of Data Mining B = WQ whereas prediction of the responses can be obtained from PCA is a linear transformation of a large number of interrelated variables into a new set of variables, called as the principal Y XWQ components (PCs), which are uncorrelated and ordered so that the first few retain most of the variation present in all the original Although PLSR is not specifically designed to discriminate variables [132]. among groups, Barker and Rayens [135] have demonstrated that PLSR can be used for such purposes by connecting PLSR and Linear Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Discriminant Analysis (LDA); this combined method is called as = 1, 2, …, K). The decomposition of both the predictor and the Partial Least Square Discriminant Analysis (PLS-DA). In PLS-DA, response block based on N-PLS model are as follows group membership is transformed into a dummy matrix, and this dummy matrix provides the response variables for PLSR. J K X  T W W  E ijk  ic jc kc ijk c1 Y  V V  F il ic lc il c1 The array X is decomposed into a tri-linear model consisting of one score vector for observation called tc (I x 1), and two weight vectors, one for type I variable called w (J x 1) and one for type II variable called w (K x 1). Similarly, a bi-linear model is used in decomposing the matrix Y into one score vector vc (I x 1) and one weight vector uc (L x 1). The decomposition is conducted such that the covariance among the score of predictor t and the corresponding score of the response v is maximized. All scores and weights are indexed with c showing that they correspond to cth multiway component, while C represents the total number of multiway components used in N-PLS model. Moreover, E and F are the residuals of the decomposition of the three-dimensional array X and matrix Y, respectively. 5 Figure 2. Schematic diagram of the decomposition of both predictor and Figure 3. Illustration of matricizing three-dimensional array X (I x J x K) into response blocks for: (a) PLS and (b) N-PLS model. matrix X (I x JK). Furthermore, let Xk (I x J) be the kth slice of X (I x J x K) for the corresponding kth of type II variable, then matricizing three- An extension of PLSR to deal with multidimensional data known dimensional array X into matrix X (I x JK) is performed as follows as Multiway Partial Least Squares has been developed by Bro [136] [137] and is called as N-PLS. In this model, the same principle of PLSR for two dimensional data is utilized, that is, both predictor and response X = [X | X | … | X ] 1 2 K blocks are decomposed successively into multi-linear model such that the pairwise scores have maximal covariance. The score of the Fig. 3 depicts this unfolding process of array X into matrix X. predictor is then regressed to the response variable. Fig. 2 illustrates Using this notation, the score tc of the cth component can be the decomposition of N-PLS model. Moreover, N-PLS model can calculated as [138] also be used for discrimination purpose, which is called as N-PLS- DA, that is the multiway version of PLS-DA, by utilizing the dummy K J matrix of group membership as the response variable. t  X(w w ) c c c Consider the three-dimensional array X indexed by observation (i or = 1, 2, …, I), type I variable (j = 1, 2, …, J) and type II variable (k Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics pharmacological activities: A2 and A4. Plant P2 also has two J K J K pharmacological activities: A1 and A2, while plant P3 has three t  x w w ic  ijk jc kc activities: A3, A4, and AK. The other connections can be described j1 k1 similarly. From the concept of integrated platform of knowledge of From Eq. (12), the weight corresponding to cth component, wc medicinal plants and plant and human-omics depicted in Fig. 1, the (JK x 1), can be defined as efficacy layer in Fig. 4 represents the physiological activity layer in human-omics attribute, the herbal medicine and plant layer represent K J the prescription and medicinal herb layer, respectively, in knowledge w  (w w ) c c c of medicinal plants attribute, while the pharmacological activity layer represents the metabolomics layer in plant-omics attribute. On the Smilde [140] also described that, due to the deflation in X during following section we will illustrate the data mining techniques on the decomposition, the weight matrix W (JK x C) can be applied herbal medicine database analyzing relationship among entities for directly to the original unfolded matrix X is defined as two, and more than two attributes. t t t t W [w | (I w w )w | ... | (I w w )(I w w )...(I w w )w ] 2 JK 1 1 2 JK 1 1 JK 2 2 JK Q1 Q1 Q Hence, the scores in T (I x C) expressed directly in terms of the X- columns is T = XW After the decomposition procedure, the next step is to regress Y on the component scores T Y  TB with t -1 t B = (T T) T Y From Eq. (15) and (16) we have Figure 4. A typical network illustrating connections between efficacy, herbal medicine, plant, and pharmacological activity of plant. Y XWB Therefore, the regression coefficients BNPLS (JK x L) needed to predict Y from X are obtained as As an illustration for data mining of herbal medicine database which rely on relationship between two attributes, the relationship B = WB NPLS between the efficacy of Jamu and medicinal plants used in Jamu is explored using PCA [143-145]. The efficacies of 3,138 Jamu are 4. Illustration of Data Mining Techniques classified into one of nine categories, namely: (1) disorders of appetite (DOA), (2) disorders of mood and behavior (DMB), (3) female Indonesia, the mega-biodiversity center like Brazil, has at least reproductive organ problems (FML), (4) gastrointestinal disorders 9,600 species of plants with pharmacological activity [110] and has (GST), (5) musculoskeletal and connective tissue disorders (MSC), developed blended herbal medicines called Jamu taking modern and (6) pain/inflammation (PIN), (7) respiratory disease (RSP), (8) traditional knowledge of herbs into consideration. To prepare Jamu, urinary related problems (URI), and (9) wounds and skin infections several plants are selected and mixed such that the concoction has the (WND). In total, those 3,138 Jamu use 465 plants in their desired efficacy. Traditionally, plants are chosen based on prior ingredients. The distribution of Jamu and plant utilized in Jamu for experience which is passed down from generation to generation. In each efficacy is shown in Table 2. curing a particular disease, each ethnic group in Indonesia may have Note that, one plant may be used in many Jamu with varying its own formulas, whose specific nature depends strongly on the local efficacies. Hence, it is interesting to find out the most significant plant resources in the region where a given population lives and the effects of specific plants by analyzing their usage in Jamu, and efficacies of Jamu medicines have been empirically demonstrated considering that the more useful a given plant in having certain effect, [139-142]. Data mining techniques with the blended herbal medicine the more frequently the plant will be used in Jamu when that effect is databases such as KAMPO and JAMU (Fig. 1) makes it possible to desired. Biplot, a multivariate exploration tool, is suitable for this comprehensively and mathematically understand those blended herbal purpose because it provides simultaneous plot of principal component systems. Fig. 4 illustrates a network connecting efficacy, herbal scores and loadings, as representation of observations and variables, medicine, plant, and pharmacological activity of plant. The network respectively [145]. Considering plants as observations and efficacy showing that crude medicines M1, which is useful for efficacy E1, use groups as variables, the relationship between them can be explored three plants in its ingredients: plant P1, P3, and P4. Plant P1 has two using a biplot. Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics corresponding efficacy (see Table 2). Efficacies with large variability of plants usage (MSC, GST, and FML) have large values for both factors; in contrast, efficacies with small variability of plants usage (efficacy DMB, URI, and RSP) have small values for both factors. In the configurations, many plants are clustered in the center. Note that, the projection value of plants' point on a given efficacy line is the prediction of the frequency of plants usage on that efficacy. So, these clustered plants are basically plants whose frequencies of usage in Jamu are very low. In contrast to the clustered plants, some plants are spread out and located near the efficacy for which the plants are highly utilized. For example, Ginger (Zingiber officinale) is located near the efficacy MSC. Ginger is well known for its function of refreshing body, and for this reason many Jamu use Ginger for efficacy MSC which can easily be identified from biplot configuration. Another example is Turmeric (Curcuma longa) which located near the efficacy FML. Due to its analgesic and antimicrobial activity, this plant is well known and highly utilized in Indonesia as ingredient of Jamu formula for women during menstruation, which is a problem that classified into efficacy FML. Thus, the biplot configuration exhibits useful information in exploring the relationship between plants and the efficacy of Jamu. Another illustration for relationship between two attributes on Following the explanation of PCA in previous section, the data data mining of herbal medicine database is the modeling of Jamu matrix A as an input for PCA is generated by putting plant as ingredients (representation of knowledge of medicinal plants) to observation and efficacy as variables. So, A consists of 465 rows and 9 predict the efficacy (representation of human omics). This analysis is columns. Each cell aij shows the number of Jamu that use plant i and performed because of the fact that Jamu is prepared from a mixture of useful for efficacy j. several plants. The plants are chosen so that the Jamu has the desired efficacy. As a result, the composition of the plants used in Jamu formula determines the efficacy. Thus, it is interesting to model the ingredients of Jamu, i.e. the constituent plants, and use this model to predict efficacy. PLS-DA, a statistical model for classification and discrimination based on Partial Least Square Regression (PLSR), is suitable for this analysis because a large number of plants are used in Jamu, whereas Jamu efficacies can be grouped into a few categories or classes. In this method, the plants used in each Jamu medicine served as the predictors, whereas the efficacy of each Jamu provided the responses. The data structure used for PLS-DA is as follows. The data matrix X in X-block contains plant usage status. The dimension of matrix X is (I x J), where I is the number of Jamu (in this case, 3,138), and J is the number of plants (in this case, 465). Because of the availability of information about Jamu products, which generally do not state in detail the mixing ratio of the plants used, the predictors X is constructed only in binary data. Each cell xij (i = 1, 2, …, I; j = 1, 2, …, J) is set to 1 if Jamu i uses plant j, and is set to 0 otherwise. In the present study, nine indicator variables, which correspond to the 9 efficacies listed in Table 2 perform as the Y- block in PLS-DA modeling. Thus, the dimension of data matrix Y is Figure 5. Biplot configuration based on PCA analysis of Jamu data. Plants and Jamu efficacies are represented as red points and blue lines, (I x 9). Each cell yil (l = 1, 2, …, 9) is set to 1 if Jamu i is classified respectively. into efficacy group l, and is set to 0 otherwise. Note that because each Jamu is classified to one efficacy only. y  1  il l1 Biplot configuration using the first two components is shown in Using the derived PLS-DA model, we can then use it to predict Fig. 5. In the figure, plants are represented as red points while Jamu the efficacy of Jamu given information of the ingredients. In this efficacies as blue lines, i.e. vectors based on loadings. The length of a analysis, among the 3,138 Jamu medicines, the efficacies of 2,248 given efficacy line showing the variability of plant usage for the Jamu medicines (71.6%) can be assigned to an individual efficacy corresponding efficacy, that is, the longer the efficacy line the larger reported. Hence, the efficacy in most Jamu medicines can be predicted the variability of plant usage for that efficacy. From Fig. 6, it is on the basis of medicinal plants used. The percentages of correct obvious that efficacy MSC has the largest variability of plant usage, prediction for each efficacy (see Table 3) vary from 22.7% for followed by efficacy GST and FML. On the other hand, efficacy efficacy DMB to 89.8% for efficacy GST. The low percentage of DMB has the smallest variability of plant usage, followed by efficacy correct prediction for efficacy DMB can be addressed due to the small URI and RSP. This finding can be addressed due to two factors, that number of Jamu for this efficacy, which is only 22 out of 3,138 Jamu is, the number of Jamu as well as the number of plant utilized in the (see Table 2). Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Furthermore, plants in the ingredients of Jamu are used as main the PLS-DA coefficients obtained from this process generates a ingredients, which contribute primarily to the medicines' efficacies; distribution, against which a p-value can be calculated and other plants are used as supporting ingredients [146, 147]. subsequently evaluated for significance [150]. Investigating which plants are main ingredients and which are The results of the significance testing of all plants used in each 9 supporting is important in order to comprehensively understand the efficacies are shown in Table 4. Note that one plant may be used for mechanisms by which specific plants achieve desired efficacies. The more than one efficacy. From the testing, we observed 234 plants regression coefficients of previous PLS-DA model, which relates (50.3% among all 465 plants) showing no significant status for all 9 plants usage in Jamu as predictors and Jamu efficacy as response, can efficacies; whereas the other 231 plants have significant status which be helpful in this attempt because they summarize the effect of plant comprise of 189 plants (40.6%) are significant only for 1 efficacy, 38 on efficacy. Plants that act as main ingredients will have significant plants (8.2%) are significant for 2 efficacies, and the other 4 plants effect on the model developed. Furthermore, due to the absence of (0.9%) are significant for 3 efficacies. Besides testing the plants usage parametric testing for the PLS-DA coefficients, the evaluation for statistically, furthermore, we also checked from scientific papers the significance is performed using permutation testing, in which the usage of significant plants in their corresponding efficacy. Many of distribution of coefficients under the null hypothesis is generated via the results we obtained by our analysis are supported by scientific resampling of the existing data [149]. papers. Note that in predicting Jamu efficacy based on the information of its ingredients we can also use other methods such as discrimination analysis, nominal logistic regression, and support vector machine. However, in the present study we focus on PLS-DA in classifying Jamu efficacy by taking into consideration that we also intend to evaluate the significance of plant usage in Jamu to achieve specific efficacy as well as extending the analysis into three-way model by adding the plant pharmacological activity into predictors’ block. Figure 6. Clustergram of pharmacological activity against Jamu efficacy. The red and black cells indicate that the pharmacological activity is significant or non-significant, respectively, to the corresponding efficacy. The resampling is performed by permuting the order of the responses (in this case, Jamu efficacies) while maintaining the order of the predictors (in this case, plant utilization as Jamu ingredients) so that the existing relationship between the predictors and the response is destroyed and a new data set is generated under the null hypothesis, i.e., plant utilization in Jamu does not affect Jamu efficacy. If we perform such resampling many times and apply the PLS-DA model on the new data generated from the resampling, the accumulation of Volume No: 4, Issue: 5, January 2013, e201301010 Data Mining Methods for Omics harmful to health, then it is reasonable that antimicrobial activity is important and should be available in many Jamu formulas in During the modeling process of PLS-DA in the previous section, Indonesia. It should be noted that many popular medicinal plants in the ingredients of Jamu provide the predictor while the Jamu efficacy Indonesia such as Temulawak (Curcuma xanthorriza), Ginger serves as the response. In order to identify the function of the plants (Zingiber officinale), Turmeric (Curcuma longa) or Kencur in Jamu to achieve specific efficacy, the reported pharmacological (Kaempferia galanga) have content of this activity [152]. activities of the plants are added to the predictors block. Thus, the Anti-inflammation, antispasmodic, analgesic, sedative, and predictors block can be represented as a three-dimensional array X (I stimulant are also clustered into this general activity group. Since x J x K) indexed by Jamu medicine (i), plant (j), and pharmacological many health problems or diseases are often accompanied with activity (k) as depicted in Fig. 2 with Jamu medicine, plant, and inflammation or spasm, then the plants with anti-inflammation pharmacological activity serve as observation, type I and type II and/or antispasmodic activity are chosen in many Jamu formulas. variables, respectively. Furthermore, the response block is represented Those health problems/diseases often cause pain or other as matrix Y (I x 9). This analysis then connects three attributes: (1) discomforts, thus plants with certain activities such as analgesic or knowledge of medicinal plants (represented by Jamu and plants sedative effects are chosen in many Jamu medicines. Finally, stimulant corresponding to JAMU DB in Fig 1); (2) plant omics (represented activity, which excites or quickens activity of the physiological by pharmacological activity corresponding to Biological activity (Nat) processes, is important for the recovery reason after one experiencing in Fig 1); and (3) human omics (represented by efficacy). those health problems or diseases. The detail about the elements of array X and matrix Y is as the From the previous explanation regarding the grouping of following. Let xijk (k = 1, 2, …, K; K = 46 where K is the number of pharmacological activity, it can be concluded that in formulating Jamu reported pharmacological activity; see previous section on definition the plants are selected so that, beside curing the targeted diseases or of i, j, I, and J) denotes the usage status of plant j with health problems as indicated by the specific activities, the plants also pharmacological activity k in Jamu i, where xijk = 1 if the plant j with should overcome the other discomforts caused by the targeted diseases pharmacological activity k is used in Jamu i, and xijk = 0 otherwise. or health problems as indicated by the general activities. It is in On the other hand, let yil represents the status of Jamu i on efficacy l, accordance with the process of making the Jamu medicines that where yil = 1 if Jamu i is classified into efficacy l, and yil = 0 involving whole part of plant and not only the specific active otherwise. components. Hence specific or general pharmacological activities of In order to identify the pharmacological activity that is components are involved during the curing process of Jamu medicines significantly related with the efficacy, we adopt the guidelines from towards targeted diseases or health problems. Hair et al. [150] that all weights w (in absolute values) of 0.3 or above are significant for sample sizes of 350 or greater. Figure 6 depicts the 2-dimensional dendrogram of Jamu efficacy and the pharmacological activity significantly related with the efficacy. The 5. Concluding Remarks cluster of Jamu efficacy and the pharmacological activity was performed using Ward Linkage based on the Euclidean distance Biology, like most scientific disciplines, is in an era of accelerated among the entities. The clustering of the pharmacological activity side information gathering and scientists increasingly depend on the clearly exhibits two groups. The first group consists of activities availability of amounts of data such as nucleotide and protein useful for one or two efficacies only. This group can be regarded as a sequences, protein and gene expression, dynamics of metabolites etc. group of specific activity because the effects of the activities are The nature of current systematic understanding of big data biology specific for certain efficacy. For example the diuretic activity is useful towards health, nutrition, and other societal issues have recently for efficacy URI and DOA. Diuretic is an agent that increases the become the focus of scholar in societal studies of science and secretion and elimination of urine from the body [151]. Obviously, information studies. The rise of community databases, i.e., this activity is beneficial for the efficacy URI. Diuretic also help the KNApSAcK family DB introduced in the present review, has been body eliminate waste and support the whole process of inner strongly associated with the current emphasis on data-intensive cleansing, which is an action that is useful for efficacy DOA especially science. The central question is whether scientists can deduce how related with a slimming purpose. The five activities systems and whole organisms work from this torrent of molecular (antihaemorrhoidal, carminative, hypoglycaemic, depurative, and data. To progress this situation, data-intensive approach is needed for anthelmintic) are specifically related with efficacy GST. understanding intra- and inter-relations in individual layers Antihaemorrhoidal means an activity that treats haemorrhoids (piles), represented in Fig. 1. The former can be solved based on a type of while the carminative is defined as an activity that eases discomfort multivariate analyses such as cluster analysis and principal component caused by flatulence. Hypoglycaemic activity helps reduce the levels of analysis. Though the latter is more complicated, several approaches sugar in the blood, whereas the depurative eliminates toxins and including PLS and N-PLS make it possible to clarify and understand purifies the system especially the blood, and the anthelmintic helpful those relations. The big data biology has become an inevitable part of in expelling parasites from the gut. Thus, all of these activities are biology, and the laws of nature could be clarified based on global helpful for the problem related with the digestive system, i.e. the analysis of big data biology the era of which has appeared. For efficacy GST. centuries biological research mainly depended on experiments and for Furthermore, the second group of activity revealed by the a decade or two computational analysis has usually followed dendrogram consists of activities useful for at least four efficacies. In experimentation but future it might be the opposite i.e., contrast to the first group, this group can be regarded as the general computational analysis is done first to guide the experimental design activities because of the diverse efficacies related to this group. Among facilitated by versatile and freely available omics data at various all activities clustered to this group, antimicrobial activity is databases. significantly related with all 8 efficacies. We can interpret this result as follows. Due to the environmental conditions, hygiene, and its location as a tropical country which led to many microbes that are Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Competing Interests: The authors have declared that no competing interests exist. © 2013 Afendi et al. Licensee: Computational and Structural Biotechnology Journal. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are properly cited. What is the advantage to you of publishing in Computational and Structural Biotechnology Journal (CSBJ) ? Easy 5 step online submission system & online manuscript tracking Fastest turnaround time with thorough peer review Inclusion in scholarly databases Low Article Processing Charges Author Copyright Open access, available to anyone in the world to download for free WWW.CSBJ.ORG Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Computational and Structural Biotechnology Journal Pubmed Central

Data Mining Methods for Omics and Knowledge of Crude Medicinal Plants toward Big Data Biology

Loading next page...
 
/lp/pubmed-central/data-mining-methods-for-omics-and-knowledge-of-crude-medicinal-plants-zXzC3VJfMl

References (263)

Publisher
Pubmed Central
Copyright
© Afendi et al.
ISSN
2001-0370
eISSN
2001-0370
DOI
10.5936/csbj.201301010
Publisher site
See Article on Publisher Site

Abstract

Volume No: 4, Issue: 5, January 2013, e201301010, http://dx.doi.org/10.5936/csbj.201301010 CSBJ Data Mining Methods for Omics and Knowledge of Crude Medicinal Plants toward Big Data Biology a,b a a d c a a Farit M. Afendi , Naoaki Ono , Yukiko Nakamura , Kensuke Nakamura , Latifah K. Darusman , Nelson Kibinge , Aki Hirai Morita , Ken e f a a,* Tanaka , Hisayuki Horai , Md. Altaf-Ul-Amin , Shigehiko Kanaya Abstract: Molecular biological data has rapidly increased with the recent progress of the Omics fields, e.g., genomics, transcriptomics, proteomics and metabolomics that necessitates the development of databases and methods for efficient storage, retrieval, integration and analysis of massive data. The present study reviews the usage of KNApSAcK Family DB in metabolomics and related area, discusses several statistical methods for handling multivariate data and shows their application on Indonesian blended herbal medicines (Jamu) as a case study. Exploration using Biplot reveals many plants are rarely utilized while some plants are highly utilized toward specific efficacy. Furthermore, the ingredients of Jamu formulas are modeled using Partial Least Squares Discriminant Analysis (PLS-DA) in order to predict their efficacy. The plants used in each Jamu medicine served as the predictors, whereas the efficacy of each Jamu provided the responses. This model produces 71.6% correct classification in predicting efficacy. Permutation test then is used to determine plants that serve as main ingredients in Jamu formula by evaluating the significance of the PLS-DA coefficients. Next, in order to explain the role of plants that serve as main ingredients in Jamu medicines, information of pharmacological activity of the plants is added to the predictor block. Then N-PLS-DA model, multiway version of PLS-DA, is utilized to handle the three-dimensional array of the predictor block. The resulting N-PLS-DA model reveals that the effects of some pharmacological activities are specific for certain efficacy and the other activities are diverse toward many efficacies. Mathematical modeling introduced in the present study can be utilized in global analysis of big data targeting to reveal the underlying biology. 1. Introduction Data-intensive sciences have progressed in modern astronomy [1], the rapid increasing of omics data produced by genomics, biology [2-8], computational materials science [9], ecology [10-11] transcriptomics, proteomics and metabolomics [2-8]. This situation and social science [12] because open-access data has increased is also a feature of the ethnomedicinal survey and the number of drastically. Data-intensive or -driven discovery in biology requires a medicinal plants is estimated to be 40,000 to 70,000 around the large open pool of data across the full breadth of the life sciences and world [13] and many countries utilize these plants as blended herbal the access to the pool will invite “New” logic, strategies and tools to medicines, e.g., China (traditional Chinese medicine), Japan (Kampo discover new trends, associations, discontinuities, and exceptions that medicine), India (Ayruveda, Siddha and Unani) and Indonesia reveal aspects of the underlying biology [2, 5, 6]. Big data biology, (Jamu). Blended herbal medicines as well as single herb medicines which is a discipline of data-intensive science, was proposed based on include a large number of constituent substances which exert effects on human physiology through a variety of biological pathways. To comprehensively understand the medicinal usage of plants based upon traditional and modern knowledge, we add to KNApSAcK Family database systems the selected herbal ingredients i.e., the formulas of Graduate School of Information Science, Nara Institute of Science and Kampo and Jamu, omics information in plants and humans, and Technology, Nara 630-0101, Ikoma, Japan b physiological activities in humans [14-16]. These information need to Department of Statistics, Bogor Agricultural University, Jln. Meranti, be connected in a way that enables scientists to make predictions Kampus IPB Darmaga, Bogor 16680, Indonesia based on general principles. Biopharmaca Research Center, Bogor Agricultural University, Kampas IPB In this mini-review, we discuss the usage of KNApSAcK Family Taman Kencana, Jln. Taman Kencana No. 3 Bogor 16151, Indonesia DB in metabolomics, explain mining techniques such as principal Maebashi Institute of technology, 450-1 Kamisadori, Maebashi-shi, component analysis (PCA), partial least square regression (PLSR) and Gunma, 371-0816 Japan multiway model, and show their application on Indonesian blended Department of Medicinal Resources, Institute of Natural Medicine, herbal medicines (Jamu) as a case study. University of Toyama, 2630 Toyama, 930-0194, Japan Department of Electronic and Computer Engineering, Ibaraki National 2. KNApSAcK Family Database College of Technology, 866 Nakane, Hitachinaka, Ibaraki 312-8508, Japan Omics biology, like most scientific disciplines, is in an era of * Corresponding author. accelerated increase of data, so called big data biology [2-8]. Large- E-mail address: skanaya@gtc.naist.jp (Shigehiko Kanaya) scale sequencing centers, high-throughput analytical facilities and Data Mining Methods for Omics individual laboratories produce vast amounts of data such as planning to translate them into English as early as possible. Lunch nucleotide and protein sequences, gene expression measurements, Box DB comprises information on 800 edible species which include protein and genetic interactions, mass spectra of metabolites and the species introduced to Japan from outside or originally grown in phenotype studies. The goal of investigating the interactions between Japan, general information of the crops and the effect of them on medicinal/edible plants and humans is to comprehensively understand human health. the molecular mechanism of medicinal plants on human physiology Noncommunicable diseases such as heart disease, metabolic based on current and traditional knowledge. Optimization of blended disease, cancer and respiratory disease, which superseded the herbal formulas should be developing using information derived from infectious diseases because of the development and widespread plant and human omics. To reach this goal we need to develop distribution of vaccines and antimicrobial drugs, account for 60% of databases based on the platform shown in Fig. 1A. KNApSAcK all deaths worldwide and 80% of deaths in low- and middle-income family DBs have been developed for this purpose [14-16]. Relations countries [17]. Food and ingredients in sanative diet and more among individual DBs are illustrated in Fig. 1A and main page of effective combination of foods beneficial against those KNApSAcK Family DB is shown in Fig. 1B. noncommunicable diseases are accumulated in DietNavi and DietDish DBs, respectively (b and d in Fig. 1). FoodProcessor DB comprises 309 retortable pouch foods encompassed by 261 food ingredients produced in Japan, and connected with DietNavi and KNApSAcK core by species names of foods. To systematize crude drugs by multifaceted view points, we have developed four DBs (WorldMap, KAMPO, JAMU and TeaPot DBs as shown in e-h of Fig. 1). The KNApSAcK WorldMap DB comprises 46,256 geographic zone-plant pair entries in 217 geographical zones except mini-states such as the Principalities of Liechtenstein, Monaco and Andorra, and the Vatican City. Prescriptions corresponding to Japanese and Indonesian herbal medicines have been accumulated in KAMPO and JAMU DBs, respectively. KAMPO DB is comprised of 1,581 primary formulas classified in to 336 formula names encompassed by 278 medicinal plants which are approved by the National health insurance authority in Japan. JAMU DB is comprised of 5,310 formulas encompassed by 550 medicinal plants and 12 anatomical regions which are approved by the National Agency of Drug and Food Control (NA-DFC) of Indonesia. Medicinal/edible plants reported in the scientific literature have been classified into geographic zones using the International Organization for Standardization (ISO3166), which defines geographic zones based on the borders between nations and small islands. Herbs are defined as any plants with leaves, seeds, and flowers used for flavoring, food, medicine, perfume and parts of such a plant as used in cooking. Those are accumulated in TeaPot DB. Two types of biological activities, that is, activities of natural resources and metabolites to other species including human, i.e., antibiotic, anticancer and so on are accumulated in Natural Activity and Metabolite Activity DBs (Fig. 1B), respectively. The former and the latter comprised 33,703 and 6,677 entries, respectively. For extension of species-metabolite relationship DB to metabolic pathways, it is needed to design secondary metabolic pathway DB for detection of metabolic pathways based on enzyme reactions and prediction of reactions by peptide sequences. So we have developed Motoercycle DB containing 2,421 entries. The metabolomics of plants is developing rapidly [18-20 and references in Table 1], and it will be an important topic in the systems-biological studies of interactions between plants and humans, which is included in the topics of big data biology [2-8], with the goal of achieving a holistic understanding of plant function and healthcare, including the activity Figure 1. Integrated platform of knowledge of medicinal plants and plant of medicinal plants as well as interaction between plants and their and human –omics and KNApSacK Family databases. (A) The relations of environment [14-16, 21, 22]. attributes among individual DBs. (B) Main window of KNApSAcK Family To facilitate access to metabolite information obtained from DB, indexes from a to i in panel A correspond to those in panel B. analytical techniques, we have developed species-metabolite relationship DB (KNApSAcK Core DB) which contains 106,418 species-metabolite relationships encompassing 21,705 species and Four DBs (Lunch Box DB, DietNavi DB, Food Processor DB and 50,897 metabolites. Nine databases of KNApSAcK family (except DietDish DB, a-d in Fig. 1) are about Food & Health related with DietDish) are connected with KNApSAcK Core DB to easily obtain Japanese foods and ingredients explained in Japanese language because candidates of secondary metabolites in species utilized in several initially we developed them targeting the Japanese people, but we are purposes [23]. The KNApSAcK Core DB was utilized in very Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics diverged purposes of metabolomics studies including identification of Consider a data matrix A = (a1 a2 … ap) with n observations and metabolites (‘Exp’ in Table 1), construction of integrated databases let V (p x p) be the variance-covariance matrix of A. The principal (‘DB’), bioinformatics and systems biology (‘Bioinfo’) , and cited in at components of A, Z = (z1 z2 … zp), are calculated as least 110 papers listed in Table 1, that is, in 29 papers in the period of 2006-2008, 25 papers in the period of 2009, 20 papers in 2010, z = Ac ( j = 1, 2, …, p) j j 18 papers in 2011, 18 papers in 2012-2013. In addition, it was applied in diverged species from bacteria to plants and animals, in where cj is the j-th eigenvector of V which correspond to the j-th total 28 species, that is, Angelica acutiloba [74], Arabidopsis lyrata eigenvalue of V (j). The properties of PCs are: (1) Var(zj) = j; (2) ssp. petraea [56], Arabidopsis thaliana[25, 30, 33, 35, 37, 46, 47, 62, Cov(zj,zj’) = 0, j  j’; (3) Var(z1)  Var(z2)  … Var(zp). The 70, 86, 99, 103, 104, 108, 109, 121, 122], Atriplex halimus [127], cumulative proportion of variance of the original variables explained Bacillus subtilis [113], Brassica oleraceae var capitata [60], Brufelsia by the first J principal components can be obtained as calycina [81], Capsicum sp. [123], Citrus sinensis [131], Curcuma longa [77], Ephedra sp. [67], Escherichia coli [51], Fragaria x ananassa [40, 43, 44], Fragaria vesca [105], Glycine max [53],   j j1 Glycyrrhiza uralensis [94], Hordeum vulgare [80, 102], Homo Pr(z ) sapiens [63, 101], Jatropha curcas [124, 125], Malx x domestica  j [126], Ophiorrhiza pumila [117], Oryza sativa [49, 61], Papaver j1 somniferum [42], Rattus norvegicus [39, 97], Rizotania solani [79], Solanum lycopersicum [45, 48], Solanum tuberosum [98] and Zea mays [120]. PLSR is a regression method, which assumes underlying factors In the period of 2006-2008, many review papers [‘Review’ in among the predictors account for most of the response variation [133, Table 1] focused on metabolomics platforms integrated by mass- 134]. These underlying factors of X-variate spectrometry and metabolite databases including KNApSAcK Core [29, 31, 34, 38, 42, 49, 52] and on linking chemistry with biology T = XW [24], and on metabolome researches targeting the model plant Arabidopsis thaliana [30, 33, 35, 37]. In 2009, metabolome studies are obtained by maximizing their covariance with the corresponding were extended to diverged species such as crops and medicinal plants underlying factors of Y-variate where X is an n  m matrix of [53, 60, 61, 67, 68, 73, 74, 78] and to engineering studies such as quality assessment based on metabolomics [73, 74]. Thus predictors, Y is an n  p matrix of responses, T is an n  c matrix of metabolomics was applied from model species to crops and medicinal X-score factors, and W is m  c matrix of weight. Note that n is the herbs. In the period of 2010-2013, metabolomics was further number of observations, m is the number of predictors, p is the extended to genetics such as QTL [80, 98, 126], and to explanation number of responses, and c is the number of components. of species by metabolites, i.e., ecological subjects [85] phytoalexins The X-score factors, i.e. matrix T, have the following [119], herbivore-induced metabolites [120] and defense against properties [133]. pathogens [131], and to stress responses [115, 116, 127]. In addition, metabolomics has also been tried in imaging studies [112, 129]. a. When multiplied by loadings P, they are good summaries of X, Species-metabolite relation database KNApSAcK Core has been i.e. the X-residuals E are small utilized in the extended fields of metabolomics researches and the horizon of metabolomics researches could be recognized by reviewing X = TP + E the works that utilized and/or cited the KNApSAcK DB. Methodologies for multivariate analysis to statistically process the b. The X-score factors are good predictors of Y, i.e. massive amount of metabolome data were reviewed in [16] and to systematize blended herbal medicines in Kampo [15]. In the following Y = TQ + F section, we focus on the mining studies of blended herbal medicines for systematically understanding the composition of medicinal herbs The Y-residuals F express the deviations between the observed and to efficacies on humans, that is, principal component analysis (PCA) modeled responses. that makes it possible to systematize the ingredient in individual Based on Eq. (3), Eq. (5) can be rewritten as a multiple regression blending systems, partial least squares (PLS) that can relate the model ingredients of medicinal herbs to the efficacies and N-PLS that can connect multi-factors to the efficacies. We initially explain individual Y = XWQ + F = XB + F techniques in Section 3 and then discuss their application in data- mining of blended types of herbal medicines in Section 4. Thus, PLSR coefficients B can be written as 3. Mathematical Methods of Data Mining B = WQ whereas prediction of the responses can be obtained from PCA is a linear transformation of a large number of interrelated variables into a new set of variables, called as the principal Y XWQ components (PCs), which are uncorrelated and ordered so that the first few retain most of the variation present in all the original Although PLSR is not specifically designed to discriminate variables [132]. among groups, Barker and Rayens [135] have demonstrated that PLSR can be used for such purposes by connecting PLSR and Linear Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Discriminant Analysis (LDA); this combined method is called as = 1, 2, …, K). The decomposition of both the predictor and the Partial Least Square Discriminant Analysis (PLS-DA). In PLS-DA, response block based on N-PLS model are as follows group membership is transformed into a dummy matrix, and this dummy matrix provides the response variables for PLSR. J K X  T W W  E ijk  ic jc kc ijk c1 Y  V V  F il ic lc il c1 The array X is decomposed into a tri-linear model consisting of one score vector for observation called tc (I x 1), and two weight vectors, one for type I variable called w (J x 1) and one for type II variable called w (K x 1). Similarly, a bi-linear model is used in decomposing the matrix Y into one score vector vc (I x 1) and one weight vector uc (L x 1). The decomposition is conducted such that the covariance among the score of predictor t and the corresponding score of the response v is maximized. All scores and weights are indexed with c showing that they correspond to cth multiway component, while C represents the total number of multiway components used in N-PLS model. Moreover, E and F are the residuals of the decomposition of the three-dimensional array X and matrix Y, respectively. 5 Figure 2. Schematic diagram of the decomposition of both predictor and Figure 3. Illustration of matricizing three-dimensional array X (I x J x K) into response blocks for: (a) PLS and (b) N-PLS model. matrix X (I x JK). Furthermore, let Xk (I x J) be the kth slice of X (I x J x K) for the corresponding kth of type II variable, then matricizing three- An extension of PLSR to deal with multidimensional data known dimensional array X into matrix X (I x JK) is performed as follows as Multiway Partial Least Squares has been developed by Bro [136] [137] and is called as N-PLS. In this model, the same principle of PLSR for two dimensional data is utilized, that is, both predictor and response X = [X | X | … | X ] 1 2 K blocks are decomposed successively into multi-linear model such that the pairwise scores have maximal covariance. The score of the Fig. 3 depicts this unfolding process of array X into matrix X. predictor is then regressed to the response variable. Fig. 2 illustrates Using this notation, the score tc of the cth component can be the decomposition of N-PLS model. Moreover, N-PLS model can calculated as [138] also be used for discrimination purpose, which is called as N-PLS- DA, that is the multiway version of PLS-DA, by utilizing the dummy K J matrix of group membership as the response variable. t  X(w w ) c c c Consider the three-dimensional array X indexed by observation (i or = 1, 2, …, I), type I variable (j = 1, 2, …, J) and type II variable (k Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics pharmacological activities: A2 and A4. Plant P2 also has two J K J K pharmacological activities: A1 and A2, while plant P3 has three t  x w w ic  ijk jc kc activities: A3, A4, and AK. The other connections can be described j1 k1 similarly. From the concept of integrated platform of knowledge of From Eq. (12), the weight corresponding to cth component, wc medicinal plants and plant and human-omics depicted in Fig. 1, the (JK x 1), can be defined as efficacy layer in Fig. 4 represents the physiological activity layer in human-omics attribute, the herbal medicine and plant layer represent K J the prescription and medicinal herb layer, respectively, in knowledge w  (w w ) c c c of medicinal plants attribute, while the pharmacological activity layer represents the metabolomics layer in plant-omics attribute. On the Smilde [140] also described that, due to the deflation in X during following section we will illustrate the data mining techniques on the decomposition, the weight matrix W (JK x C) can be applied herbal medicine database analyzing relationship among entities for directly to the original unfolded matrix X is defined as two, and more than two attributes. t t t t W [w | (I w w )w | ... | (I w w )(I w w )...(I w w )w ] 2 JK 1 1 2 JK 1 1 JK 2 2 JK Q1 Q1 Q Hence, the scores in T (I x C) expressed directly in terms of the X- columns is T = XW After the decomposition procedure, the next step is to regress Y on the component scores T Y  TB with t -1 t B = (T T) T Y From Eq. (15) and (16) we have Figure 4. A typical network illustrating connections between efficacy, herbal medicine, plant, and pharmacological activity of plant. Y XWB Therefore, the regression coefficients BNPLS (JK x L) needed to predict Y from X are obtained as As an illustration for data mining of herbal medicine database which rely on relationship between two attributes, the relationship B = WB NPLS between the efficacy of Jamu and medicinal plants used in Jamu is explored using PCA [143-145]. The efficacies of 3,138 Jamu are 4. Illustration of Data Mining Techniques classified into one of nine categories, namely: (1) disorders of appetite (DOA), (2) disorders of mood and behavior (DMB), (3) female Indonesia, the mega-biodiversity center like Brazil, has at least reproductive organ problems (FML), (4) gastrointestinal disorders 9,600 species of plants with pharmacological activity [110] and has (GST), (5) musculoskeletal and connective tissue disorders (MSC), developed blended herbal medicines called Jamu taking modern and (6) pain/inflammation (PIN), (7) respiratory disease (RSP), (8) traditional knowledge of herbs into consideration. To prepare Jamu, urinary related problems (URI), and (9) wounds and skin infections several plants are selected and mixed such that the concoction has the (WND). In total, those 3,138 Jamu use 465 plants in their desired efficacy. Traditionally, plants are chosen based on prior ingredients. The distribution of Jamu and plant utilized in Jamu for experience which is passed down from generation to generation. In each efficacy is shown in Table 2. curing a particular disease, each ethnic group in Indonesia may have Note that, one plant may be used in many Jamu with varying its own formulas, whose specific nature depends strongly on the local efficacies. Hence, it is interesting to find out the most significant plant resources in the region where a given population lives and the effects of specific plants by analyzing their usage in Jamu, and efficacies of Jamu medicines have been empirically demonstrated considering that the more useful a given plant in having certain effect, [139-142]. Data mining techniques with the blended herbal medicine the more frequently the plant will be used in Jamu when that effect is databases such as KAMPO and JAMU (Fig. 1) makes it possible to desired. Biplot, a multivariate exploration tool, is suitable for this comprehensively and mathematically understand those blended herbal purpose because it provides simultaneous plot of principal component systems. Fig. 4 illustrates a network connecting efficacy, herbal scores and loadings, as representation of observations and variables, medicine, plant, and pharmacological activity of plant. The network respectively [145]. Considering plants as observations and efficacy showing that crude medicines M1, which is useful for efficacy E1, use groups as variables, the relationship between them can be explored three plants in its ingredients: plant P1, P3, and P4. Plant P1 has two using a biplot. Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics corresponding efficacy (see Table 2). Efficacies with large variability of plants usage (MSC, GST, and FML) have large values for both factors; in contrast, efficacies with small variability of plants usage (efficacy DMB, URI, and RSP) have small values for both factors. In the configurations, many plants are clustered in the center. Note that, the projection value of plants' point on a given efficacy line is the prediction of the frequency of plants usage on that efficacy. So, these clustered plants are basically plants whose frequencies of usage in Jamu are very low. In contrast to the clustered plants, some plants are spread out and located near the efficacy for which the plants are highly utilized. For example, Ginger (Zingiber officinale) is located near the efficacy MSC. Ginger is well known for its function of refreshing body, and for this reason many Jamu use Ginger for efficacy MSC which can easily be identified from biplot configuration. Another example is Turmeric (Curcuma longa) which located near the efficacy FML. Due to its analgesic and antimicrobial activity, this plant is well known and highly utilized in Indonesia as ingredient of Jamu formula for women during menstruation, which is a problem that classified into efficacy FML. Thus, the biplot configuration exhibits useful information in exploring the relationship between plants and the efficacy of Jamu. Another illustration for relationship between two attributes on Following the explanation of PCA in previous section, the data data mining of herbal medicine database is the modeling of Jamu matrix A as an input for PCA is generated by putting plant as ingredients (representation of knowledge of medicinal plants) to observation and efficacy as variables. So, A consists of 465 rows and 9 predict the efficacy (representation of human omics). This analysis is columns. Each cell aij shows the number of Jamu that use plant i and performed because of the fact that Jamu is prepared from a mixture of useful for efficacy j. several plants. The plants are chosen so that the Jamu has the desired efficacy. As a result, the composition of the plants used in Jamu formula determines the efficacy. Thus, it is interesting to model the ingredients of Jamu, i.e. the constituent plants, and use this model to predict efficacy. PLS-DA, a statistical model for classification and discrimination based on Partial Least Square Regression (PLSR), is suitable for this analysis because a large number of plants are used in Jamu, whereas Jamu efficacies can be grouped into a few categories or classes. In this method, the plants used in each Jamu medicine served as the predictors, whereas the efficacy of each Jamu provided the responses. The data structure used for PLS-DA is as follows. The data matrix X in X-block contains plant usage status. The dimension of matrix X is (I x J), where I is the number of Jamu (in this case, 3,138), and J is the number of plants (in this case, 465). Because of the availability of information about Jamu products, which generally do not state in detail the mixing ratio of the plants used, the predictors X is constructed only in binary data. Each cell xij (i = 1, 2, …, I; j = 1, 2, …, J) is set to 1 if Jamu i uses plant j, and is set to 0 otherwise. In the present study, nine indicator variables, which correspond to the 9 efficacies listed in Table 2 perform as the Y- block in PLS-DA modeling. Thus, the dimension of data matrix Y is Figure 5. Biplot configuration based on PCA analysis of Jamu data. Plants and Jamu efficacies are represented as red points and blue lines, (I x 9). Each cell yil (l = 1, 2, …, 9) is set to 1 if Jamu i is classified respectively. into efficacy group l, and is set to 0 otherwise. Note that because each Jamu is classified to one efficacy only. y  1  il l1 Biplot configuration using the first two components is shown in Using the derived PLS-DA model, we can then use it to predict Fig. 5. In the figure, plants are represented as red points while Jamu the efficacy of Jamu given information of the ingredients. In this efficacies as blue lines, i.e. vectors based on loadings. The length of a analysis, among the 3,138 Jamu medicines, the efficacies of 2,248 given efficacy line showing the variability of plant usage for the Jamu medicines (71.6%) can be assigned to an individual efficacy corresponding efficacy, that is, the longer the efficacy line the larger reported. Hence, the efficacy in most Jamu medicines can be predicted the variability of plant usage for that efficacy. From Fig. 6, it is on the basis of medicinal plants used. The percentages of correct obvious that efficacy MSC has the largest variability of plant usage, prediction for each efficacy (see Table 3) vary from 22.7% for followed by efficacy GST and FML. On the other hand, efficacy efficacy DMB to 89.8% for efficacy GST. The low percentage of DMB has the smallest variability of plant usage, followed by efficacy correct prediction for efficacy DMB can be addressed due to the small URI and RSP. This finding can be addressed due to two factors, that number of Jamu for this efficacy, which is only 22 out of 3,138 Jamu is, the number of Jamu as well as the number of plant utilized in the (see Table 2). Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Furthermore, plants in the ingredients of Jamu are used as main the PLS-DA coefficients obtained from this process generates a ingredients, which contribute primarily to the medicines' efficacies; distribution, against which a p-value can be calculated and other plants are used as supporting ingredients [146, 147]. subsequently evaluated for significance [150]. Investigating which plants are main ingredients and which are The results of the significance testing of all plants used in each 9 supporting is important in order to comprehensively understand the efficacies are shown in Table 4. Note that one plant may be used for mechanisms by which specific plants achieve desired efficacies. The more than one efficacy. From the testing, we observed 234 plants regression coefficients of previous PLS-DA model, which relates (50.3% among all 465 plants) showing no significant status for all 9 plants usage in Jamu as predictors and Jamu efficacy as response, can efficacies; whereas the other 231 plants have significant status which be helpful in this attempt because they summarize the effect of plant comprise of 189 plants (40.6%) are significant only for 1 efficacy, 38 on efficacy. Plants that act as main ingredients will have significant plants (8.2%) are significant for 2 efficacies, and the other 4 plants effect on the model developed. Furthermore, due to the absence of (0.9%) are significant for 3 efficacies. Besides testing the plants usage parametric testing for the PLS-DA coefficients, the evaluation for statistically, furthermore, we also checked from scientific papers the significance is performed using permutation testing, in which the usage of significant plants in their corresponding efficacy. Many of distribution of coefficients under the null hypothesis is generated via the results we obtained by our analysis are supported by scientific resampling of the existing data [149]. papers. Note that in predicting Jamu efficacy based on the information of its ingredients we can also use other methods such as discrimination analysis, nominal logistic regression, and support vector machine. However, in the present study we focus on PLS-DA in classifying Jamu efficacy by taking into consideration that we also intend to evaluate the significance of plant usage in Jamu to achieve specific efficacy as well as extending the analysis into three-way model by adding the plant pharmacological activity into predictors’ block. Figure 6. Clustergram of pharmacological activity against Jamu efficacy. The red and black cells indicate that the pharmacological activity is significant or non-significant, respectively, to the corresponding efficacy. The resampling is performed by permuting the order of the responses (in this case, Jamu efficacies) while maintaining the order of the predictors (in this case, plant utilization as Jamu ingredients) so that the existing relationship between the predictors and the response is destroyed and a new data set is generated under the null hypothesis, i.e., plant utilization in Jamu does not affect Jamu efficacy. If we perform such resampling many times and apply the PLS-DA model on the new data generated from the resampling, the accumulation of Volume No: 4, Issue: 5, January 2013, e201301010 Data Mining Methods for Omics harmful to health, then it is reasonable that antimicrobial activity is important and should be available in many Jamu formulas in During the modeling process of PLS-DA in the previous section, Indonesia. It should be noted that many popular medicinal plants in the ingredients of Jamu provide the predictor while the Jamu efficacy Indonesia such as Temulawak (Curcuma xanthorriza), Ginger serves as the response. In order to identify the function of the plants (Zingiber officinale), Turmeric (Curcuma longa) or Kencur in Jamu to achieve specific efficacy, the reported pharmacological (Kaempferia galanga) have content of this activity [152]. activities of the plants are added to the predictors block. Thus, the Anti-inflammation, antispasmodic, analgesic, sedative, and predictors block can be represented as a three-dimensional array X (I stimulant are also clustered into this general activity group. Since x J x K) indexed by Jamu medicine (i), plant (j), and pharmacological many health problems or diseases are often accompanied with activity (k) as depicted in Fig. 2 with Jamu medicine, plant, and inflammation or spasm, then the plants with anti-inflammation pharmacological activity serve as observation, type I and type II and/or antispasmodic activity are chosen in many Jamu formulas. variables, respectively. Furthermore, the response block is represented Those health problems/diseases often cause pain or other as matrix Y (I x 9). This analysis then connects three attributes: (1) discomforts, thus plants with certain activities such as analgesic or knowledge of medicinal plants (represented by Jamu and plants sedative effects are chosen in many Jamu medicines. Finally, stimulant corresponding to JAMU DB in Fig 1); (2) plant omics (represented activity, which excites or quickens activity of the physiological by pharmacological activity corresponding to Biological activity (Nat) processes, is important for the recovery reason after one experiencing in Fig 1); and (3) human omics (represented by efficacy). those health problems or diseases. The detail about the elements of array X and matrix Y is as the From the previous explanation regarding the grouping of following. Let xijk (k = 1, 2, …, K; K = 46 where K is the number of pharmacological activity, it can be concluded that in formulating Jamu reported pharmacological activity; see previous section on definition the plants are selected so that, beside curing the targeted diseases or of i, j, I, and J) denotes the usage status of plant j with health problems as indicated by the specific activities, the plants also pharmacological activity k in Jamu i, where xijk = 1 if the plant j with should overcome the other discomforts caused by the targeted diseases pharmacological activity k is used in Jamu i, and xijk = 0 otherwise. or health problems as indicated by the general activities. It is in On the other hand, let yil represents the status of Jamu i on efficacy l, accordance with the process of making the Jamu medicines that where yil = 1 if Jamu i is classified into efficacy l, and yil = 0 involving whole part of plant and not only the specific active otherwise. components. Hence specific or general pharmacological activities of In order to identify the pharmacological activity that is components are involved during the curing process of Jamu medicines significantly related with the efficacy, we adopt the guidelines from towards targeted diseases or health problems. Hair et al. [150] that all weights w (in absolute values) of 0.3 or above are significant for sample sizes of 350 or greater. Figure 6 depicts the 2-dimensional dendrogram of Jamu efficacy and the pharmacological activity significantly related with the efficacy. The 5. Concluding Remarks cluster of Jamu efficacy and the pharmacological activity was performed using Ward Linkage based on the Euclidean distance Biology, like most scientific disciplines, is in an era of accelerated among the entities. The clustering of the pharmacological activity side information gathering and scientists increasingly depend on the clearly exhibits two groups. The first group consists of activities availability of amounts of data such as nucleotide and protein useful for one or two efficacies only. This group can be regarded as a sequences, protein and gene expression, dynamics of metabolites etc. group of specific activity because the effects of the activities are The nature of current systematic understanding of big data biology specific for certain efficacy. For example the diuretic activity is useful towards health, nutrition, and other societal issues have recently for efficacy URI and DOA. Diuretic is an agent that increases the become the focus of scholar in societal studies of science and secretion and elimination of urine from the body [151]. Obviously, information studies. The rise of community databases, i.e., this activity is beneficial for the efficacy URI. Diuretic also help the KNApSAcK family DB introduced in the present review, has been body eliminate waste and support the whole process of inner strongly associated with the current emphasis on data-intensive cleansing, which is an action that is useful for efficacy DOA especially science. The central question is whether scientists can deduce how related with a slimming purpose. The five activities systems and whole organisms work from this torrent of molecular (antihaemorrhoidal, carminative, hypoglycaemic, depurative, and data. To progress this situation, data-intensive approach is needed for anthelmintic) are specifically related with efficacy GST. understanding intra- and inter-relations in individual layers Antihaemorrhoidal means an activity that treats haemorrhoids (piles), represented in Fig. 1. The former can be solved based on a type of while the carminative is defined as an activity that eases discomfort multivariate analyses such as cluster analysis and principal component caused by flatulence. Hypoglycaemic activity helps reduce the levels of analysis. Though the latter is more complicated, several approaches sugar in the blood, whereas the depurative eliminates toxins and including PLS and N-PLS make it possible to clarify and understand purifies the system especially the blood, and the anthelmintic helpful those relations. The big data biology has become an inevitable part of in expelling parasites from the gut. Thus, all of these activities are biology, and the laws of nature could be clarified based on global helpful for the problem related with the digestive system, i.e. the analysis of big data biology the era of which has appeared. For efficacy GST. centuries biological research mainly depended on experiments and for Furthermore, the second group of activity revealed by the a decade or two computational analysis has usually followed dendrogram consists of activities useful for at least four efficacies. In experimentation but future it might be the opposite i.e., contrast to the first group, this group can be regarded as the general computational analysis is done first to guide the experimental design activities because of the diverse efficacies related to this group. Among facilitated by versatile and freely available omics data at various all activities clustered to this group, antimicrobial activity is databases. significantly related with all 8 efficacies. We can interpret this result as follows. Due to the environmental conditions, hygiene, and its location as a tropical country which led to many microbes that are Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org Data Mining Methods for Omics Competing Interests: The authors have declared that no competing interests exist. © 2013 Afendi et al. Licensee: Computational and Structural Biotechnology Journal. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are properly cited. What is the advantage to you of publishing in Computational and Structural Biotechnology Journal (CSBJ) ? Easy 5 step online submission system & online manuscript tracking Fastest turnaround time with thorough peer review Inclusion in scholarly databases Low Article Processing Charges Author Copyright Open access, available to anyone in the world to download for free WWW.CSBJ.ORG Volume No: 4, Issue: 5, January 2013, e201301010 Computational and Structural Biotechnology Journal | www.csbj.org

Journal

Computational and Structural Biotechnology JournalPubmed Central

Published: Mar 23, 2013

There are no references for this article.