Protoplasma DOI 10.1007/s00709-016-0953-3
ORIGINAL ARTICLE
Artificial neural network-based model for the prediction of optimal growth and culture conditions for maximum biomass accumulation in multiple shoot cultures of Centella asiatica Archana Prasad 1 & Om Prakash 2 & Shakti Mehrotra 1 & Feroz Khan 2 & Ajay Kumar Mathur 1 & Archana Mathur 1
Received: 26 September 2015 / Accepted: 3 February 2016 # Springer-Verlag Wien 2016
Abstract An artificial neural network (ANN)-based modelling approach is used to determine the synergistic effect of five major components of growth medium (Mg, Cu, Zn, nitrate and sucrose) on improved in vitro biomass yield in multiple shoot cultures of Centella asiatica. The back propagation neural network (BPNN) was employed to predict optimal biomass accumulation in terms of growth index over a defined culture duration of 35 days. The four variable concentrations of five media components, i.e. MgSO4 (0, 0.75, 1.5, 3.0 mM), ZnSO4 (0, 15, 30, 60 μM), CuSO4 (0, 0.05, 0.1, 0.2 μM), NO3 (20, 30, 40, 60 mM) and sucrose (1, 3, 5, 7 %, w/v) were taken as inputs for the ANN model. The designed model was evaluated by performing three different sets of validation experiments that indicated a greater similarity between the target and predicted dataset. The results of the modelling experiment suggested that 1.5 mM Mg, 30 μM Zn, 0.1 μM Cu, 40 mM NO3 and 6 % (w/v) sucrose were the respective optimal concentrations of the tested medium components for achieving maximum growth index of 1654.46 with high centelloside yield (62.37 mg DW/culture) in the cultured multiple shoots. This Handling Editor: Peter Nick * Archana Mathur
[email protected];
[email protected]
1
Division of Plant Biotechnology, CSIR-Central Institute of Medicinal & Aromatic Plants (CSIR-CIMAP); PO CIMAP, Lucknow 226015, India
2
Division of Molecular and Structural Biology, CSIR-Central Institute of Medicinal & Aromatic Plants (CSIR-CIMAP); PO CIMAP, Lucknow 226015, India
study can facilitate the generation of higher biomass of uniform, clean, good quality C. asiatica herb that can efficiently be utilized by pharmaceutical industries. Keywords Centella asiatica . Multiple shoot cultures . Media optimization . Artificial neural network modelling
Introduction Biological processes are highly complex and dynamic, and are often influenced by genetic and environmental factors. These highly variable factors are largely responsible for nondeterministic and non-linear nature of developmental processes of biological entities. To understand these complex and unpredictable developmental phenomenons of biological systems, artificial neural network (ANN) modelling has emerged as an efficient tool (Gallego et al. 2010; PérezPiñeiro et al. 2012). Artificial neural modelling utilizes and processes a set of multi-dimensional data as inputs and produce outputs indicating the relationship between the data (Nazmul et al. 1997; Patnaik 1999). Processing of such data through ANN thus leads to the output that can predict a biological response under the synergistic influence of several variables. Due to this property, the ANN-based modelling is applicable to a wide area of scientific advances including health, agro-technology, genetic engineering, and plant biotechnology (Shao et al. 2006; Jiménez et al. 2008; Landín et al. 2009; Gago et al. 2010a). In plant biotechnology, with particular reference to plant tissue culture research, the ANN-based calculation for optimizating an in vitro process/protocol offers better alternatives for predicting a
A. Prasad et al.
response under various growth conditions (Mehrotra et al. 2008; Mehrotra et al. 2013; Zielinska and Kepczynska 2013; Alanagh et al. 2014). Since plant tissue cultures are generally used to manipulate tissue growth to enhance biomass yield in lesser time or to derive useful metabolites/ compounds by various approaches, optimization of physical and nutritional parameters under in vitro conditions is a regular practice. In such conventional approach for optimizing the growth conditions, alteration in concentration of various medium components is individually done at a time, and the effect on the tissue growth is chased for each variable. Assessment of synergistic effect of various culture components simultaneously is a serious missing link, in such an approach that makes the effort insufficient, tedious, unreliable and challenging in terms of time, labour and cost. To overcome these challenges, ANN-based computational scheming is advantageous as it can be efficiently utilized for the prediction of combined effect of various independent variables and select the best combination that would lead to the optimized results. Centella asiatica is an important anti-Alzheimer, antiinflammatory and neuroprotective medicinal herb, used both in traditional as well as modern systems of medicine. The high therapeutic value of Centella derived from saponins (asiaticoside and madecassoside) and sapogenins (asiatic acid and madecassic acid) (Gohil et al. 2010; Roy et al. 2013) predominantly present in the aerial parts of the plant. The plant is also widely used in wound healing and scar management caused by diseases like leprosy, psoriasis and keloid (Widgerow 2000; Lee et al. 2012; Somboonwong et al. 2012; Zahara et al. 2014). In keloid treatment, the herb is also used to suppress collagen expression and TGF-β/Smad signalling through inducing Smad 7 and inhibiting TGF βRI and TGF βRII in keloid fibroblasts (Tang et al. 2011). Antibacterial, fungicidal, antioxidant and anticancer activities have also been ascribed to C. asiatica (Bridgman 2003; Arumugam et al. 2011). Presently, the bulk of the market demand of C. asiatica herb is fulfilled by its collection from the wild populations. Since the natural stands of C. asiatica mainly grow in marshy lands, the collected herb is highly contaminated by heavy metals, microbial loads and insoluble ash contents. These constraints have made tissue culturebased cultivation of Centella an attractive tool for the production of good quality, clean herb in large amounts. In vitro propagation of C. asiatica for biomass production has been tested by several workers (Bhandari et al. 2013; Tiwari et al. 2013; Singh et al. 2014). In our laboratory also, method for raising Centella biomass in tissue culture and hydroponic setup has been attempted (Prasad et al. 2012a, b). Few publications on superimposing the influence of biotic/abiotic elicitors on growth and centelloside accumulation in Centella tissue culture have also appeared (Kim et al. 2004; Prasad et al. 2013). In most of these reported studies, the effects of different
growth and nutrients factors were assessed individually, and the level of biomass/metabolite productivity never reached to a commercially attractive scale. Therefore, in the present study, an effort has been made to apply ANN modelling approach to assess if synergistic influence of five different growth nutrients be predicted for obtaining more optimized biomass production. Earlier also, Omar et al. (2005) have applied a statistical model to elucidate growth of C. asiatica cell suspension culture. This study therefore constitutes the first report on application of ANN model for optimization and prediction of biomass yield in multiple shoot cultures of C. asiatica.
Material and methods BPNN A feed-forward back propagation type network was created for input layer (five input nodes) with a single hidden layer (three nodes) and output layer with one node (Fig. 1). Optimization of model was done by using different parameters like one thousand epochs as maximum, momentum and learning rate (1), output layer learning rate (0.3), initial weight ± range (0.5) and data normalization between 0.1 and 0.9. The variable concentrations of five media components, i.e. MgSO4 (mM), ZnSO4 (μM), CuSO4 (μM), NO3 (mM) and sucrose (%w/v), were the inputs to the ANN model. Each input is associated with a weight (w), which gets modified during the training of the model. The neuron set computes the weighted sum of its inputs through sigmoid function. ! X yi ¼ f wi j yi þ b j
The output (yi) of one neuron provides the input for the other neuron. The weighted sum was the net input to the neuron. The wij refers to the weight from neuron j to neuron i. The bias value to the neuron is indicated by b. The neuronal function can be varied according to the optimization need of the model. The output neuron provided the net output from the input neurons as growth index (GI). Algorithm of training of artificial neural network employs gradient descent, using back propagation to compute the actual gradients for local minima. The whole ANN was optimized (based on learning rate and
Fig. 1 ANN architecture for the development of model
Artificial neural network-based model for the prediction Table 1
Statistical validation of model
Training/test ratio
Samples in external test set
Regression coefficient for test set (R2test)
90:10
10
0.966
80:20 70:30
20 30
0.960 0.951
60:40
40
0.953
50:50
50
0.961
momentum for learning) to achieve global minima for facilitating the best simulated model based on provided dataset. Statistical validation of model The model was statistically validated using regression coefficient (R2), self-cross validation regression coefficient (Q2), 10fold cross-validation regression coefficient (R210-fold) and LOO regression coefficient (R2LOO). Besides, performance of the model was also evaluated with five different ratios of training: test sets (Table 1). Plant material, media compositions and culture conditions In vitro multiple shoot cultures of C. asiatica were established from nodal segments of C. asiatica plants as previously described (Prasad et al. 2012a). The micro-shoots were maintained in liquid Murashige and Skoog’s medium (1962) (10 ml per 100 ml Erlenmeyer flask) containing 2.5 mg/l kinetin and 3 % (w/v) sucrose. The pH of the medium was adjusted to 5.8 prior to autoclaving (121 °C, 1 kg/cm2/s for 15 min). All cultures were maintained under white, cool (40 μmol/m2/s) fluorescent light for 16-h photoperiod at 25 ± 3 °C and 60–70 % relative humidity (RH). Stock cultures were sub-cultured after every 35 days onto the fresh medium with similar composition.
at 1, 3, 5 and 7 % (Prasad et al. 2012a). The rest of the medium composition and the physical culture conditions remained the same as described above in all experimental setups. The shoots were cultured under these conditions for 35 days. After 35 days of culture, the shoots from each treatment were harvested and observed for growth enhancement (output) in terms of growth index (GI) that was calculated as increase in final fresh weight over the initial inoculum weight. The experiment was repeated thrice with five replicates per treatment each. Chemical extraction and HPLC analysis For chemical analysis, 100 mg of lyophilized dried leaf tissue powder was extracted for analysing their respective centelloside constituents as reported earlier (Prasad et al. 2014). In brief, the lyophilized sample was extracted thrice with 80 % methanol (5 ml) for 48 h and then filtered through Whatman filter paper no 1. De-fatting of these methanolic fractions was done using n-hexane (1:1) and the resultant extracts were concentrated in rotavapour (BUCHI, R-210; Switzerland) for HPLC analysis. Waters HPLC system (Waters, Miliford, USA) with C18 column (150 mm × 4.6 mm i.d.; 3.5 μm), 600 E pump, 2996 photo diode array detector, 717 auto-sampler was used for analysing the four centellosides namely madecassoside, asiaticoside, madecassic acid and asiatic acid. A gradient elution programming was performed for quantitative and qualitative analysis using two solvents i.e. solvent A [H2O: acetonitrile (ACN): methanol (MeOH): acetic acid (AA):: 70:10:20:0.15, v/v/v/v] and solvent B (H2O: ACN: MeOH: AA:: 10:50:40:0.15, v/v/v/v) as mobile phase. A linear gradient programming at 27 °C was carried out where the initial composition of the solvent was 100 % A, changing to 80 % A at 5 min, and 10 % A at 25 min, while a constant flow rate of 1.0 ml/min was maintained up to 25 min. After 30 min, the flow rate was changed to 1.2 ml/min while maintaining the solvent composition at 10 % A. After 35 min, the initial
Experimental designs and data procurement 2000 1800 1600
y = 0.953x + 88.974 R² = 0.9765
1400 Predicted GI
The tissue culture experiments were performed in MS basal growth medium. Variables were given in five major components of the media viz. MgSO4 (mM), ZnSO4 (μM), CuSO4 (μM), NO3 (mM) and sucrose (% w/v). These variables were taken as inputs in the ANN model developed for the study. The concentrations of these components were optimized in the medium to improve the biomass yield of multiple shoot cultures. Individual experiments were performed to assess the effect of individual component at four different levels on multiple shoot cultures of C. asiatica. The micro-/macro- nutrients varied in the experimental design as MgSO4 0, 0.75, 1.5 and 3 mM, ZnSO4 0, 15, 30 and 60 μM, CuSO4 0, 0.05, 0.1 and 0.2 μM, NO3 20, 30, 40 and 60 mM, and sucrose level varied
1200 Training set
1000 800
Test set (50%)
600
Linear (Training set)
400 200 0
0
500
1000
1500
2000
Experimental GI
Fig. 2 Regression plot between growth index obtained from tissue culture experimental set-ups versus growth index predicted through ANN-based model
A. Prasad et al. Table 2
Growth index (GI) obtained through experimental setups, model prediction and validation experiments data
Experiment set MgSO4 (mM) ZnSO4 (μM) CuSO4 (μM) NO3 (mM) Sucrose (%) Experimental growth index
Model predicted Experimental validated growth index growth index
1
1.16
30
0.1
28.89
2.55
400 to 539.86
474.39
466.61
2
1.62
27
0.09
39.72
2.83
501 to 935.83
645.10
631.54
3 Control
1.5 1.5
30 30
0.1 0.1
40 40
6 3
1667 to 1745.79 1764.45 587 to 796 693.72
optimal shoot growth in terms of growth index (GI) under in vitro conditions was predicted through this model. The success of plant tissue culture cycle to obtain a desired in vitro response is significantly influenced by a synergistic action of many culture variables. Among these, the levels of macro-/micro-nutrients and carbon source present in the culture medium plays the most important role in the manifestation of growth and cellular functions in vitro. The expression of resultant final response, however, represents the sum of synergistic interplay and effect of these medium components. The present study highlights the synergistic influence of five nutrient factors on growth and biomass accumulation in multiple shoot cultures of C. asiatica by using a back propagation neural network model. The model was found suitable for predicting the optimal shoot growth under an optimized set-up of these nutritional variables. Details of the back propagation
conditions were restored until the completion of the run (50 min). The peaks were identified using the authentic standards of asiaticoside, madecassoside, asiatic acid and madecassic acid at λ max 206. The reference standards of asiatic acid, asiaticoside, madecassoside and madecassic acid were bought from Fluka Analytical, France, Sigma-Aldrich, USA, and Chroma Dex Ltd., USA, respectively.
Results and discussion BPNN model assessment and validation In the present study, an effort has been made to elucidate the combined effect of five different media components on in vitro shoot growth of C. asiatica with the use of back propagation neural network model. The resultant Table 3 Comparison of growth indices (GI) of C. asiatica in vitro shoot cultures in different tissue culture experiments and GI predicted from ANN-based model
1654.46 587.10
MgSO4 (mM)
Sucrose (%)
ZnSO4 (μM)
CuSO4 (μM)
NO3 (mM)
Tested GIa
0
3
30
0.1
40
423
507.368
0.75 1.5
3 3
30 30
0.1 0.1
40 40
648 587
679.994 693.720
3 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5
3 1 3 5 7 3 3 3 3 3 3 3 3 3 3 3 3
30 30 30 30 30 0 15 30 60 30 30 30 30 30 30 30 30
0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0 0.05 0.2 0.1 0.1 0.1 0.1
40 40 40 40 40 40 40 40 40 40 40 40 40 20 30 40 60
590 501 601 1756 1699 747 661 796 550 590 783 730 871 400 631 587 612
623.644 539.862 693.720 1745.790 1732.910 815.213 739.158 693.720 670.622 693.720 848.749 752.337 935.828 459.550 631.682 693.720 691.440
a
Each value is the average of five replicates
Predicted GI
Artificial neural network-based model for the prediction
Fig. 3 Multiple shoot cultures of C. asiatica grown under three models predicted experimental conditions
neural network (BPNN) model assessment and validation are described below. The employed regression model was evaluated on the basis of four parameters: regression coefficient (R2) = 0.976703, self-cross validation regression coefficient (Q2) = 0.951296, 10-fold cross-validation regression coefficient (R 2 10f o l d ) = 0 . 9 3 4 3 5 2 a n d L O O r e g r es s i on c o e ff i c i e n t (R2LOO) = 0.963704. The regression plot shows efficient prediction for test set also (Fig. 2). The model has been reevaluated with different set of training: test set ratio of 90:10, 80:20, 70:30, 60:40 and 50:50. Regression coefficients for test set (R2test) were found to be ranging between 0.966, 0.960, 0.951, 0.953 and 0.961, respectively. The crossvalidation results indicated the suitability of the model for prediction in real-time culture environment. The statistically validated model was subsequently used to design three experiments with different datasets. These three sets of experiments were performed with different range of input variables (Table 2). Multiple shoot cultures grown under the three predicted experimental sets were harvested after 35 days of culture cycle for the measurement of their respective fresh weights biomass gains. These fresh weight values were further used to calculate the GI in each experimental set. Different combinations of micro-/macro-nutrients along with carbon source lead to significant outputs in three different experimental sets. Growth indices (GI) of three experimental sets with five different levels of input variables were compared with the Table 4
In vitro shoot biomass and centelloside production profile of C. asiatica under the three ANN-based experimental set-ups
Experimental set
1 2 3 Control
GI as predicted by the BPNN model values. The ANN model expels high correlation between experimental and predicted values (Tables 2 and 3). The experimental set 3 with 1.5 mM MgSO4, 30 μM ZnSO4, 0.1 μM CuSO4, 40 mM NO3 and 6 % sucrose recorded maximum growth index (GI = 1654.46) which was in agreement with the range of model predicted value with GI 1667 to 1764.45 (Table 2; Fig. 3). Though the ANN model developed here was mainly aimed for predicting optimal growth and biomass accumulation in C. asiatica multiple shoots, it was also found equally suitable to predict the productivity level of four major centellosides (namely asiaticoside, madecassoside, asiatic acid and madecassic acid) present in these shoots. In the experimental set 3, maximum biomass production (FW = 6.86 g; DW 2.12 g) also contributed towards highest centelloside yield of 62.37 mg DW/ culture as evident from Table 4. On the basis of the model used in the present study, it was observed that the increased level of sucrose (6 %) leads to improved GI in C. asiatica shoot cultures. Sucrose being the main carbon source is utilized for the supply of energy required for different metabolic processes of the plant under in vitro conditions. Recently, the role of sucrose as a carbon precursor and triggering molecule in various signalling pathways has also been highlighted (Baena-González et al. 2007). In a study conducted in kiwifruit micro-shoot proliferation, an ANN-based model was used to evaluate the positive influence of varying light intensities and increased exogenous supplementation of sucrose (Gago et al. 2014). The model predicted results were in agreement with the results obtained with traditional statistical procedure. In another study, a BPNN was used to evaluate optimal physical, chemical and biological culture conditions for best in vitro growth of Glycyrrhiza plants (Mehrotra et al. 2008). The predicted results were in accordance with the results obtained through experimental validation. Similarly, in the present study, the growth index (GI) of C. asiatica micropropagated shoots obtained through model prediction was highly comparable to the growth index obtained through experimental validation. Such studies advocate the use of ANN models as an efficient computational tool to predict most favourable
Mean biomass yield (g DW/culture)
0.31 ± 0.02 0.85 ± 0.11 2.12 ± 0.22 1.32 ± 0.24
Centelloside content (mg/g DW)
Centelloside yield (mg DW/culture)
Asiaticoside
Madecassoside
Asiatic acid
Madecassic acid
1.82 0.82 0.86 1.05
0.83 0.42 0.38 0.13
10.26 6.14 9.32 9.02
28.76 13.14 18.86 24.83
12.91 17.44 62.37 46.24
A. Prasad et al.
combinations of physical conditions and ions/carbon source to generate the best results in terms of shoot biomass. Omar et al. (2005) used Box-Behnken response surface model to screen the effects of three macronutrients (NO3−, NH4+ and PO43−) in C. asiatica cell suspension cultures. The model resulted in 99 % fitness to the experimental data, and the optimum concentrations of these macronutrients resulted in 16.0 g/l cell dry weight. The total triterpene in these cells were found below 4 mg/g cell dry weight. Earlier, we have also reported individual role of NH4+/NO3− ratio, concentrations of Cu2+, Mn2+, Mg2+, Zn2+, and varying levels of sucrose on growth and production of asiaticoside in in vitro grown shoot cultures (Prasad et al. 2012a). The optimized individual levels of Mg2+, Mn2+, Zn2+, Cu2+ and sucrose showed 1.12, 1.36, 1.28, 1.32 and 3-fold enhancement in GI over the control, respectively. However, in the present study, when the concentration of micro-/macro-nutrients was optimized through ANN-based computational model, their synergistic effect leads to 2.82-fold increase in GI over the control (in the experimental set 3, Table 2). Therefore, ANNbased computational scheming is an excellent option for optimizing physical, chemical and biological parameters to reduce the overall cost, time and labour-associated with a tissue culture protocol. The ANN-based back propagation networks used in the present study comprise multilayer programming that poses input and output layers which are linked by intermediate hidden layers. The hidden layers supervised various input variables through computational process and direct into an output layer in an easy way. The high correlation between the R2 value of training, test and validated dataset revealed the high efficacy of the present model used for better performance and predictability. In recent years, the computational assessment of growth phenomenon provided an easy means for modelling and understanding the role of media components (individually or in combination) involved during in vitro plant growth (Alanagh et al. 2010, 2014). The conventional methods of optimizing the growth medium without understanding the role of individual components make the practice difficult and unpredictable. For this purpose, the use of ANN-based modelling for the in vitro tissue growth is a very welcome advancement (Mehrotra et al. 2008; Gago et al. 2010b, c; Prakash et al. 2010; Alanagh et al. 2014). The use of computational/ mathematical modelling for the optimization of suitable culture conditions can reduce the overall cost, time and labour input in any micropropagation protocol and can make the in vitro protocols more suitable for commercial adoption (Prasad and Gupta 2006). In recent years, during tissue culture, the development of such strategies that results in optimum biomass production in short time duration concomitant with maximum secondary metabolite accumulation has been greatly emphasized.
Acknowledgments The authors are grateful to Director CSIR-CIMAP Lucknow for providing facilities to execute the work. A part of this work was carried out under a Department of Science & Technology project grant (DST No. SR/SO/Ps-28/07) to AM. Authors also acknowledge BSC-0121 project for model building work at CSIR-CIMAP Lucknow. Compliance with ethical standards Conflict of interest The author declares that there are no competing interests.
References Alanagh NE, Garoosi GA, Haddad R (2010) The effect of PGRs on in vitro shoot multiplication of GF677 hybrid (Prunus persica X P. amygdalus) rootstock on GNH medium. Iran J Gen Plant Breed 1: 34–43 Alanagh NE, Garoosi GA, Haddad R, Maleki S, Landín M, Gallego PP (2014) Design of tissue culture media for efficient Prunus rootstock micropropagation using artificial intelligence models. Plant Cell Tiss Org 117:349–359 Arumugam T, Ayyanar M, Pillai YJK, Sekar T (2011) Phytochemical screening and antibacterial activity of leaf and callus extracts of Centella asiatica. Bangladesh J Pharmacol 6:55–60 Baena-González E, Rolland F, Thevelein JM, Sheen J (2007) A central integrator of transcription networks in plant stress and energy signaling. Nature 448:938–943 Bhandari AK, Baunthiyal M, Bisht VK, Singh N, Negi JS (2013) A quick bud breaking response of a surface model for rapid clonal propagation in Centella asiatica (L.). Int J Biotechnol Mol Biol Res 4:93–97 Bridgman KE (2003) Herbal medicines. Faculty of Pharmacy, University of Sydney Gago J, Landín M, Gallego PP (2010a) Strengths of artificial neural networks in modeling complex plant processes. Plant Signal Behav 5:743–745 Gago J, Landín M, Gallego PP (2010b) A neuro fuzzy logic approach for modeling plant processes: a practical case of in vitro direct rooting and acclimatization of Vitis vinifera L. Plant Sci 179:241–249 Gago J, Nunez-Martinez L, Landin M, Gallego PP (2010c) Artficial neural network as an alternative to traditional statistical methodology in plant research. J Plant Physiol 167:23–27 Gago J, Núñez LM, Landín M, Flexas J, Gallego PP (2014) Modeling the effects of light and sucrose on In Vitro propagated plants: a multiscale system analysis using artificial intelligence technology. Plos one 9, e85989 Gallego PP, Gago J, Landín M (2010) Artificial neural networks technology to model and predict plant biology process. In: Suzuki K (ed) Artificial neural network methodological advances and biomedical applications. InTech, Rijeka, pp 197–216 Gohil KJ, Patel JA, Gajjar AK (2010) Pharmacological review on Centella asiatica: a potential herbal cure-all. Indian J Pharm Sci 72:546–556 Jiménez D, Pérez-Uribe A, Satizábal H, Barreto M, Van Damme P, Tomassini M (2008) A survey of neural network-based modeling in agroecology. In: Prasad B (ed) Soft computing applications in industry. STUDFUZZ 226. Springer, Berlin, pp 247–269 Kim OT, Kim MY, Hong MH, Ahnn JC (2004) Stimulation of asiaticoside accumulation in the whole plant cultures of Centella asiatica (L.) Urban by elicitors. Plant Cell Rep 23:339–344 Landín M, Rowe RC, York P (2009) Advantages of neurofuzzy logic against conventional experimental design and statistical analysis in
Artificial neural network-based model for the prediction studying and developing direct compression formulations. Eur J Pharm Sci 38:325–331 Lee JH, Kim HL, Lee MH, You KE, Kwon BJ, Seo HJ, Park JC (2012) Asiaticoside enhances normal human skin cell migration, attachment and growth in vitro wound healing model. Phytomedicine: Int J Phytother Phytopharmacol 19:1223–1227 Mehrotra S, Prakash O, Mishra BN, Dwevedi B (2008) Efficiency of neural networks for prediction of in vitro culture conditions and inoculum properties for optimum productivity. Plant Cell Tiss Org 95:29–35 Mehrotra S, Prakash O, Khan F, Kukreja AK (2013) Efficiency of neural network-based combinatorial model predicting optimal culture conditions for maximum biomass yields in hairy root cultures. Plant Cell Rep 32:309–317 Murashige T, Skoog F (1962) A revised medium for rapid growth and bio assays with tobacco tissue cultures. Physiol Plant 15:473–497 Nazmul KM, Yoshida T, Rivera SL, Saucedo VM, Eikens B, Oh GS (1997) Global and local neural network models in biotechnology: application to different cultivation process. J Ferment Bioeng 83:1–11 Omar R, Abdullah MA, Hasan MA, Marziah M, Mazlina MKS (2005) Optimization and elucidation of interaction between ammonium, nitrate and phosphate in Centella asiatica cell culture using response surface methodology. Biotechnol Bioprocess Eng 10:192–197 Patnaik PR (1999) Application of neural networks to recovery of biological products. Biotech Adv 17:477–488 Pérez-Piñeiro P, Gago J, Landín M, Gallego PP (2012) Agrobacteriummediated transformation of wheat: general overview and new approaches to model and identify the key factors involved. In: Ö zden C¸ ftc¸i Y (eds) Transgenic Plants BAdvances and Limitations^. Intech, pp 1–26 Prakash O, Mehrotra S, Krishna A, Mishra BN (2010) A neural network approach for the prediction of in vitro culture parameters for maximum biomass yields in hairy root cultures. J Theor Biol 265:570–585 Prasad VSS, Gupta SD (2006) Applications and potentials of artificial neural networks in plant tissue culture. In: Gupta SD, Ibaraki Y (eds) Plant tissue culture engineering. Springer, Berlin, pp 47–67 Prasad A, Mathur A, Singh M, Gupta MM, Uniyal GC, Lal RK, Mathur AK (2012a) Growth and asiaticoside production in multiple shoot cultures of a medicinal herb Centella asiatica L. under the influence of nutrient manipulations. J Nat Med 66:383–387
Prasad A, Pragadheesh VS, Mathur A, Srivastava NK, Singh M, Mathur AK (2012b) Growth and centelloside production in hydroponically established medicinal plant-Centella asiatica (L.). Ind Crops Prod 35:309–312 Prasad A, Mathur A, Kalra A, Gupta MM, Lal RK, Mathur AK (2013) Fungal elicitor-mediated enhancement in growth and asiaticoside content of Centella asiatica L. shoot cultures. Plant Growth Regul 69:265–273 Prasad A, Singh M, Yadav NP, Mathur AK, Mathur A (2014) Molecular, chemical and biological stability of plants derived from artificial seeds of Centella asiatica (L.) Urban—An industrially important medicinal herb. Ind Crops Prod 60:205–2011 Roy DC, Barman SK, Shaik MM (2013) Current updates on Centella asiatica: phytochemistry, pharmacology and traditional uses. Med Plant Res 3:20–36 Shao Q, Rowe RC, York P (2006) Comparison of neurofuzzylogic and neural networks in modelling experimental data of an immediate release tablet formulation. Eur J Pharm Sci 28:394–404 Singh G, Kaur B, Sharma N, Bano A, Kumar S, Dhaliwal HS, Sharma V (2014) In vitro micropropagation and cytomorphological evaluation of Centella asiatica (L.) Urban (mandukparni) from Himachal Pradesh, India ‐ an endemic, endangered and threatened herb. Plant Tissue Cult Biotechnol 24:155–171 Somboonwong J, Kankaisre M, Tantisira B, Tantisira MH (2012) Wound healing activities of different extracts of Centella asiatica in incision and burn wound models: an experimental animal study. BMC Complement Altern Med 12:103 Tang B, Zhu B, Liang Y, Bi L, Hu Z, Chen B, Zhang K, Zhu J (2011) Asiaticoside suppresses collagen expression and TGFb/Smad signaling through inducing Smad7 and inhibiting TGF-bRI and TGF-bRII in keloid fibroblasts. Arch Dermatol Res 303:563–572 Tiwari C, Bakshi M, Vichitra A (2013) A rapid two step protocol of in vitro propagation of an important medicinal herb Centella asiatica Linn. Afr J Biotechnol 12:1084–1090 Widgerow AD (2000) New innovations in Scar management. Aesth Plast Surg 24:227–234 Zahara K, Bibi Y, Tabassum S (2014) Clinical and therapeutic benefits of Centella asiatica. Pure Appl Biol 3:152–159 Zielinska A, Kepczynska E (2013) Neural modeling of plant tissue cultures: a review. Biotechnologia 94:253–268
