In Traditional Chinese Medicine

EXPERIMENTAL & INFORMATION STUDYTOPIC Information integration research on cumulative effect of 'Siqi, Wu- wei, and Guijing' in Traditional Chinese Medicine

Yang Xuming, Qi Mingyuan, Li Qian, Chen Li, Yu Zhongyi, Yang Lin aa Yang Xuming, College of Acupuncture and Massage, and repeated treatment. The slope of the regres- Shanghai University of Traditional Chinese Medicine, Shang- sion line represented the cumulative trend of the hai 201203, China effect of the herbs (β), and the standard deviation Qi Mingyuan, Li Qian, Chen Li, Yu Zhongyi, Experimental of the slope (Sβ) was compared with those of the Center for Science and Technology, Shanghai University of untreated animals (t 'test). All significantly cumula- Traditional Chinese Medicine, Shanghai 201203, China Yang Lin, Information Center, Shanghai University of Tradi- tive effect trends were applied with an artificial tional Chinese Medicine, Shanghai 201203, China neural network, which integrated the relationship Support by National Natural Science Foundation Grant among Siqi, Wuwei, and Guijing with tissues and or- (Correlation Analysis Among Herb's Properties affecting the gans. Multiple Parameters of the Blood and the Morphology of the Organism Structure, No. 81473366) RESULTS: There is a certain relationship among the Correspondence to: Prof. Yu Zhongyi, Experimental Cen- Siqi, Wuwei, Guijing and the anatomy of organs ter for Science and Technology, Shanghai University of Tradi- and tissues, but the different scores indicate that in- tional Chinese Medicine, Shanghai 201203, China. zhongy- fluence of Siqi, Wuwei, Guijing to anatomy of or- [email protected]; Prof. Yang Lin, Information Center, gans and tissues was a nonlinear state. Shanghai University of Traditional Chinese Medicine, Shang- hai 201203, China. [email protected] CONCLUSION: Results demonstrated that the ef- Telephone: +86-21-51322400; +86-21-51322048 Accepted: December 15, 2015 fects of Siqi, Wuwei, and Guijing have a morphological basis, and each concept was associated with multiple anatomical structures.

© 2016 JTCM.This is an open access article under the CC BY-NC-ND Abstract license (http://creativecommons.org/licenses/by-nc-nd/4.0/) OBJECTIVE: To study the morphological basis of the role of Siqi (cold as winter, cool as autumn, Key words: Four natures; Five flavors; Channel tro- warm as spring, hot as summer), Wuwei (five fla- pism; Morphological and microscopic findings; vors: sweet, pungent, salty, sour, and bitter), and Least-squares estimation; BP algorithm of artificial Guijing (meridian tropism) through the use of infor- neural networks mation integration.

METHODS: A 14C-2-deoxy-glucose autoradiogra- INTRODUCTION phy method was adopted to determine the overall Research into Traditional Chinese Medicine (TCM) impact of treatment with 39 herbs on functions of has continued to expand and become more in-depth, various tissues and organs. Data was measured at 4 and results have suggested that the mechanism of hs after a single dose and following the last treat- TCM does not take place along a single path.1 Regard- ment of repeated doses for a week. Least-squares less of the ingredients contained in herbs of TCM or estimation was used and fitted for each herb re- the structure of the body to respond to herbs of TCM, gression effect of organs and tissues after single all of which were multi-faceted and multi-leveled. Ex-

JTCM | www. journaltcm. com 538 August 15, 2016 |Volume 36 | Issue 4 | Yang XM et al. / Experimental & Information Study perimental research focused on Siqi, Wuwei, and Gui- quality No. SCXK (hu) 2003-0002). The study was ap- jing has taken place for some time now; and it is be- proved by the Medical Ethics Committee, Shanghai lieved that these factors are associated with body heat University of TCM (No. 09048). production processes or metabolic activities, and may also play a role in gastrointestinal organ function, the Experimental herbs liver, the nervous system, and other organs1 Results The selection principle of the TCM herbs was the com- have also suggested many indicators (multiple parame- monly used. The herbal properties were characterized ter) association with neuroendocrine activities.1 by low ambiguity with varying therapeutic functions to There is no evidence to suggest that a single physiologi- allow for rapid statistical analysis and a reduced experi- cal and biochemical index, or even several indices, mental workload. The properties of each herb have could be used to distinguish between the different roles been previously described,2-3 The TCM herbs selected of Siqi, Wuwei and Guijing. The relationships among for the present study included: Chaihu (Radix Bupleuri Siqi, Wuwei, Guijing and organs and tissues, as well as Chinensis), Chuipencao (Herba Sedi Sarmentosi), Lianq- the involved mechanisms, remain poorly understood. iao (Fructus Forsythiae Suspensae), Xuanshen (Radix Therefore, we chose 39 types of commonly used herbs Scrophulariae), Huangqin (Radix Scutellariae Baicalen- that are known to have varying indications and studied sis), Huangbai (Cortex Phellodendri Amurensis), Da- the effect of Siqi, Wuwei, and Guijing on various or- huang (Radix et Rhizoma Rhei Palmati), Fuling (Poria), gans and tissues. Muxiang (Radix Aucklandiae), Zhuling (Polyporus), In the present study, we explored the morphological Cheqianzi (Semen Plantaginis), Shanzhuyu (Fructus and structural basis of Siqi, Wuwei, and Guijing to es- Macrocarpii), Wumei (Fructus Mume), Gegen (Radix tablish the relationship between TCM concepts and Puerariae Lobatae), Wuyao (Radix Linderae Aggregatae), various organs and tissues. The results will hopefully Xiangfu (Rhizoma Cyperi), Shanzha (Fructus Crataegus promote the integration of TCM, modern medical sci- Pinnatifidae), Jineijin (Endothelium Coreneum Gigeriae ence, and information technology to further the devel- Galli), Fuzi (Radix Aconiti Lateralis Preparata), Rougui opment of TCM theory. This study adopted an infor- (Cortex Cinnamomi Cassiae), Gouteng (Ramulus Uncar- mation integration method to analyze experimental da- iae Rhynchophyllae cum Uncis), Muli (Concha Ostreae), ta and to provide an experimental basis for further ba- Jinqiancao (Herba Lysimachiae), Heshouwu (Radix sic and clinical research. Polygoni Multiflori), Roucongrong (Herba Cistanches Deserticolae), Xianmao (Rhizoma Curculiginis), Mangx- iao (Nalrii Sulfas), Yiyiren (Semen Coicis), Chuanxiong MATERIALS AND METHODS (Rhizoma Chuanxiong), Lingxiaohua (Flos Campsis), Puhuang (Pollen Typhae), Wuzhuyu (Fructus Evodiae Animals Rutaecarpae), Gaoliangjiang (Rhizoma Alpiniae Offici- CD-1CF1 mice(female ICR mice and male Balb/c nari), Zhizi (Fructus Gardeniae), Huoxiang (Herba mice, 28 day) were purchased from Sino-British Sippr/ Agastaches Rugosa), Meihua (Flos Mume), Xuanfuhua BK Lab Animal Ltd. (Shanghai, China). The animals (Flos Inulae Japonicae), Shudi (Radix Rehmanniae), and were specific pathogen-free and weighed 18-22 g. The Diyu (Radix Sanguisorbae), totaling 39 species. The animals were group-housed (n = 5 per cage) and main- herbs were purchased from Shanghai Cambridge Chi- tained under standard laboratory conditions: 12-h nese Medicine Yin Pian Factory (Shanghai, China), light-dark cycles (light on at 7:00 and off at 19:00), which covered all records of Siqi, Wuwei, and Guijing temperature range: 20-26 , relative humidity: attributes, and each property of used herbs appeared 40%-70%, food and water ad℃ libitum) (Certificate of more than five times.The attributes are listed inTable1.

Table 1 Attributes of Bupleuum and Sedum sarmentosum in 39 TCM herbs Name 1 2 3 4 5 6 7 8 9 10 11 12 Bupleurum + + Sedum sarmentosum + + + +

… Name 13 14 15 16 17 18 19 20 21 22 23 24 25 Bupleurum + + Sedum sarmentosum + +

… Notes: 1: cold; 2: heat; 3: warm; 4: cool; 5: even; 6: sweet; 7: pungent; 8: salty; 9: sour; 10: bitter; 11: odorless; 12: astringency; 13: poi- son; 14: spleen; 15: lung; 16: kidney; 17: liver; 18: heart; 19: stomach; 20: large intestine large intestine; 21: bladder; 22: bile; 23: small intestine; 24: triple energizer; 25: pericardium.

JTCM | www. journaltcm. com 539 August 15, 2016 |Volume 36 | Issue 4 | Yang XM et al. / Experimental & Information Study

Reagents and equipment 14C-2-deoxy-glucose was purchased from Amersham A Biosciences UK Limited Co. (Batch No. 54, 56). The reagents and equipment were consistent with experimental standards.1

Grouping of herbs and isotopes A total of 39 herbs were crushed into fine powder us- B ing a WK-1000A type high-speed herb grinder (Shan- dong Qing Zhou Jing Cheng Manufacturing Compa- ny Limited Co., Shandong, China), and run through a 100-mesh sieve. According to previous literature,2-4 the lower limit of the clinical dosage was selected and con- verted into a dose suitable for mice according to body C surface area. The herbal suspension was prepared according to 0.1 mL/10 g body weight. The CD-1CF1 male mice were weighed, marked, and randomly grouped. A total of 39 herb-treated groups were established, with three mice per group. The control group had ten mice. Mice from each herbal treat- Figure 1 Data acquisition ment group were fed the herbal solution by gavage, A: entire mouse slice; B: color-coded organs; C: gray value ac- The control group was treated with the same volume quisition. of distilled water using the same method. Information integration was coordinated with the com- At 4 h after the single dose, as well as the dose once a prehensive multi-source treatment information to yield day for a week, the mice were injected through the tail the data needed for an accurate and credible conclu- vein with 14C-2-deoxy-glucose (1.25 μCi/mL in a vol- sion. There are three levels of information integration ume of 0.1 mL/10 g body weight). After two hs, the which are decision level, feature level and data level. mice were sacrificed with an overdose of ethyl carba- Among them, the object of feature level integration mate (urethane) anesthesia to obtain tissue sections. was the concept information and the regional information, which was suitable for the summarized character- Data acquisition and analysis istics and rules of some problems that came from the Once the mice were sacrificed, they were skinned and original information. First, the characteristics were ex- the limbs and tail were removed. The bodies were then tracted from each information source (for example, the wrapped in aluminum foil and frozen at － 80 . slope was the cumulative effect of the herbs). Then, Once completely frozen, the bodies were removed℃ these features were integrated to obtain the evaluation from the aluminum foil and embedded in carboxy- results (such as artificial neural network integrated the methyl cellulose sodium solution, then completely fro- result). According to previous studies2,3 describing the zen at －80 and stored at －20 . Thin sagittal sec- properties of Siqi, Wuwei, and Guijing, statistical anal- tions were made℃ using a － 20 ℃sliding microtome ysis of the 39 herbs revealed the following: 1 = herb (Figure 1). ℃ had the same properties; 0 = herb did not have the The frozen sections were subjected to autoradiography same properties. In the 39 TCM herbs, the average and hematoxylin-eosin (H & E) staining, respectively. gray values resulting from treatment with the herb sus- Digital images were collected from entire sections of H pension after 4 h and 1 week (1 time/day), as well as & E-stained tissues and autoradiography X-ray film. the last dose after 4 h, were normalized. The x-axis rep- The corresponding organs and tissues were identified resents the selected time, the y-axis represents the aver- on the digital images, and the different organs and tis- age gray value at each time point, and least-squares esti- sues were color-coded, using Photoshop 7.0 software mation was used to obtain the herb slope ( 1) and (San Jose, CA, USA). The areas were overlapped with standard deviation (S 1) for each organ andβ tissue. areas on autoradiography images to automatically ob- The slope value representedβ the direction and speed tain average gray values (Figure 1) for data acquisition with which the herb affected the organs and tissues. and analysis. The average gray value data was used as Once all slopes were calculated, the slope and standard the functional state of various organs and tissues in the deviation of each herb was compared with the control control mice, as well as in the single-dose and repeat- group. If the difference was significant (P < 0.05), then ed-dose mouse groups, to calculate the cumulative ef- the value was 1 (slope was positive) or －1 (slope was fect rate (regression line slope of medication time and negative). If there was no significant difference, the val- effect). ue was 0. Data acquisition is divided into three processes in Fig- The least-squares method only was used to calculate ure 1. the slope value for the qualitative analysis of the cumu-

JTCM | www. journaltcm. com 540 August 15, 2016 |Volume 36 | Issue 4 | Yang XM et al. / Experimental & Information Study lative effect of the herbs. We did not use the nonlinear model to precisely fit the time response curve of the herbs, but only used the method of least-squares estimation to find the nonlinear change trend in the time effect of the herbs. We only performed a qualitative analysis of the herb effect on organs and tissues. In this study, The back-propagation (BP) algorithm of artificial neural networks (BP artificial neural network) could determine the classification standard according to the accepted sample similarity. This method of deter- mining mainly embodied in the network weight distri- bution. Thus, BP artificial neural network provides a good method for multi-source information integration and modeling.5 Table 2 shows the functional changes that bupleurum and hawthorn affected in different organs and tissues. Figure 2 is a mining model of the artificial neural network. In the application of neural network algorithm of Microsoft SQL Server 2012, number served as the key. Function changes of 15 different organs and tissues served as the input parameters. For example, each Figure 2 The mining model for the artificial neural network type of warm herb in Siqi affected 15 different organ in the warm Siqi herbs and tissue function changes of numerical influence. Af- Application introduction of least-squares estimation. ter comparing every warm herb slope and its standard if one of the 39 types of TCM herbs belonged to the deviation with the control group, if the difference was "warm herb," the warm value was 1; conversely, the significant (P < 0.05), then the value was 1 (slope was warm value was 0). Through the artificial neural net- positive) or －1 (negative slope); if the difference was work algorithm, we determined that the input value not significant, then the value was 0 (Table 3). impacted the value of the output value, and the "im- The output parameter was "warm" (warm herb in Siqi; pact factor" score could influence the outcome. There-

Table 2 Slope differences after comparing the bupleurum and hawthorn group with the control group Name A B C D E F G H I J K L M N O Bupleurum 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 Hawthorn 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Notes: A: brain; B: cerebellum; C: medulla; D: thymus; E: heart; F: liver; G: kidney; H: stomach; I: small intestine; G: large intestine; K: lung; L: upper spinal; M: middle spinal; N: lower spinal; O: diencephalon.

Table 3 Effect cumulative slope of WARM-related herbs compared with the control group Warm A B C D E F G H I J K L M N O Huoxiang (Herba Agastaches Rugosa) 0 0 0 0 0 1 1 1 1 0 1 1 1 1 0 Chuanxiong (Rhizoma Chuanxiong) 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Xuanfuhua (Flos Inulae Japonicae) 0 0 0 1 1 1 0 1 1 1 1 1 1 1 1 Roucongrong (Herba Cistanches 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Deserticolae) Muxiang (Radix Aucklandiae) 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 Shanzha (Fructus Crataegus 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Pinnatifidae) Shanzhuyu (Fructus Macrocarpii) 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 Dihuang (Radix Rehmanniae) 1 1 1 1 1 0 1 0 0 1 0 0 1 0 1 Wumei (Fructus Mume) 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 Heshouwu (Radix Polygoni 0 0 1 1 1 1 1 1 0 0 1 1 1 0 0 Multiflori) Wuyao (Radix Linderae Aggregatae) 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Notes: A: brain; B: cerebellum; C: medulla; D: thymus; E: heart; F: liver; G: kidney; H: stomach; I: small intestine; G: large intestine; K: lung; L: upper spinal; M: middle spinal; N: lower spinal; O: diencephalon.

JTCM | www. journaltcm. com 541 August 15, 2016 |Volume 36 | Issue 4 | Yang XM et al. / Experimental & Information Study fore, we determined the relationship between inputs herb had an inhibitory effect on the organs and tis- and outputs. The artificial neural network prediction sues). If the differences were not significant, the the val- model was a numerical variation of the function of 15 ue was assigned a 0, which indicated that the herb ex- different organs and tissues, and was measured and cal- hibited no obvious effect on the organs and tissues. culated following treatment with 39 kinds of TCM ˆ 2 ˆ 2 å(YMedication - YMedication ) + å(YNormal - YNormal ) herbs. The output layer parameter was the relationship S = n Medication + n Normal - 4 , between herbal properties in Siqi, Wuwei, and Guijing, 1 1 S = S + and the artificial neural network could identify the rela- b1Medication -b1Normal l l . tionship between the function of different organs and xxMedication xxNormal b - b t = 1Medication 1Normal tissues and Siqi, Wuwei, and Guijing. The score from S b1Medication -b1Normal ,n = n Medication + n Normal - 4. the input parameters affected the output parameters, where v is the degree of freedom. which was automatically calculated by the artificial neural network. Therefore, the relationship between the in- BP artificial neural network analysis put parameters and the output parameters could be cal- The artificial neural network is a computing model9-12 culated. that is generally used together with a physical device or The method of least-squares estimation is a standard calculation algorithm to simulate the structure and approach in regression analysis. By minimizing error function of a neural network in the organism. Its focus sum of squares, to find the minimum function match- did not completely reproduce the network of nerve ing of data, to obtain the minimum data of error sum cells in the organism by utilizing physical devices, but of squares between the calculated data and the actual extract some features available, so as to finish the com- data.That is,let the data derived from theory the most puter or other system which inconvenience to solve close to the real data.6 nonlinear problems. At present, the neural network has Cumulative effect rate for herbs ( ) developed dozens of network models. The BP artificial A set of sample observationsβ is as follows: neural network consists of an input layer, hidden layer, and output layer of three layers (hidden layer could x ,y ，x ,y x ,y ( 1 1 )( 2 2 )K ( n n ). For each, the linear regres- contain multiple layers). The hidden layer commonly Ù Ù Ù uses a sigmoid function. The BP artificial neural net- sion equation is as follows: Y = b + b x . 0 1 work uses a large number of samples for network train- def n n L = (x - x () y - y) = x y - yxn ing. The use of signal propagation from the input lay- xy å i i å i i , i=1 i=1 er hidden layer output layer forward mode of def n n L (x x )2 x 2 xn 2 transmission.→ Compared→ with the output layer error, er- xx = å i - = å i - i =1 i =1 . ror propagation adopts a reverse weight adjustment 1 n 1 n x = x y = y from the input layer direction to the output layer Among them, n å i , n å i . i =1 i =1 (back-propagation). The weight adjustment is general- Ù Ù Ù Ù L b = xy ly used in the negative gradient descent method until Therefore, b = y - x b , 1 L (1-1) 0 1 xx the input samples achieve convergence.13 Equation (1-1) is known as the least-squares estimation,7 The BP artificial neural network mathematical descrip- which is the slope of the regression line. Results indi- tion is as follows. The mathematical foundation of the cate the cumulative effect rate for herbs ( ). β BP algorithm was an order of the derivative, when the The standard deviation of the cumulative effect rate independent variables were functional dependencies. for herbs (S ) This could use the chain rule to determine the relation- The standardβ deviation of the slope 1 is calculated as: ships between changes of independent and dependent β variables. The value of error that existed between the 2 (Y - Yˆ ) 2 SSurplus å Sb1 = SSurplus = actual output data and calculated output data was an l xx , n - 2 . approximation of the neural network to the actual sys- Comparison between regression slope of treated tem. Given a set of training data, we could calculate mouse and control mouse groups the ordered derivative of error signal for an arbitrary pa- Each herb induced a quantifiable action on various tis- rameter. Then, using the gradient descent method to sues and organs. This is calculated by the difference of let the parameter change toward an order derivative the herb group regression equation slope subtracted by negative direction, the error signal was allowed to drop. 14 the control group regression equation slope. The t 'test The BP algorithm was described as follows. X can then be used to compare the difference and stan- Input: a given training set train , each training sample dard deviation with the control group. If P < 0.05, was composed of a set of inputs and a set of outputs. then the value was assigned a 1 (slope was positive and All inputs and outputs were floating-point data be- indicated that the herb had a stimulating effect on the tween [0, 1] (if not, the data had to go through a data organs and tissues). If P > 0.05, then the value was as- transformation, mapping into their [0, 1] interval). signed a －1 (slope was negative and indicated that the The structure of the neural network included the hid-

JTCM | www. journaltcm. com 542 August 15, 2016 |Volume 36 | Issue 4 | Yang XM et al. / Experimental & Information Study den layer node number, neural network each node and x1 y1 the characteristic function of the parameters. Output: The parameters of each node of the character- x2 y2 istic function in the neural networks. (1) In accordance with

+ + N l +1 N l +1 ¶ E p ¶ E p ¶fi + ,1 m ¶fl + ,1 m e l,i = = å · = å e l + ,1 m · ¶x l,i m =1 ¶x l + ,1 i ¶x l,i m =1 ¶x l,i ,we calculated the overall error for the ordered derivative xn ym formula of each parameter (function). (2) We arbitrarily chose a set data as the initial parame- Figure 3 BP artificial neural network structure ter, the general choice of (0, 0,... 0) put the initial pa- Backward error rameters as the current parameter. (3) According to the current parameter and type

2 N L E (y x ) | p = å L,M - L,M NO .Pinput m -1 , we calculated the overall error If error was small enough, put the current parame- Expected ters as output and exit; otherwise, contiue the follow- output ing steps. ¶ + E ¶ + E ¶f = å p · (4) According to the type ¶a x ÎS ¶x ¶¶ , Input Layer Hidding Layer Output Layer (where S is the set of all nodes containing a ) and the Signal stream value of the current parameter, for orderd derivative's Figure 4 Mathematical description of BP artificial neural net- value of all parameters, to calculate the overall error. work ¶ + E Da = -h (5) According to equation ¶a RESULTS k (h = (, h is the learning rate)) The artificial neural network integrated the results asso- ¶E ( )2 ciated with Siqi, Wuwei, and Guijing with organs and å¶ ¶a , tissues, as shown in Table 3, Table 4, and Table 5. we calculated the increment of each parameter and the size of the adjusted parameters. The adjusted parame- The analysis results of cumulative effect of herbs ters are the current parameters, so we return to step with Siqi (3). (Note: The h value of the selection was based on Artificial neural network mined the results between experience, which was usually relatively small at 0.01). Siqi and organs and tissues. Figure 3 is a BP artificial neural network structure. Fig- Results from Siqi-related organs and tissues are shown ure 4 is a mathematical description of the BP artificial in Table 4. The information integrated by the artificial neural network. neural network showed that the effects of Cold- and Table 4 Connection of Siqi and organs and tissues Siqi Related organs and tissues (score) Cold Cerebella (28.72) Stomach (27.84) Large intestine (14.17) Lower spinal (10.77) Heat Middle spinal (11.07) Stomach (24.03) Kidney (16.58) Small intestine (12.2) Warm Middle spinal (22.62) Upper spinal (14.41) Liver (11.76) Heart (13.06) Cool Cerebellum (10.58) Stomach (11.45) Heart (8.14) Kidney (7.41) Even Brain (46.41) Lower spinal (38.83) Small intestine (18.9) Cerebellum (16.57)

Table 5 Connection of Wuwei and organs and tissues Wuwei Related organs and tissues (score) Pungent Stomach (39.16) Small intestine (23.16) Thymus (13.73) Middle spinal (13.55) Sweet Kidney (26.74) Large intestine (14.34) Lower spinal (13.08) Cerebellum (10.59) Sour Stomach (22.32) Heart (21.1) Brain (15.03) Small intestine (10.45) Bitter Stomach (41.08) Heart (17.99) Lung (14.38) Lower spinal (12.41) Salty Kidney (28.27) Stomach (25.16) Middle spinal (22.36) Upper spinal (19.87) Astringency Upper spinal (22.76) Middle spinal (14.77) Stomach (13.27) Medulla (9.6) Odorless Kidney (14.99) Middle spinal (13.49) Upper spinal (13.25) Lung (5.5)

JTCM | www. journaltcm. com 543 August 15, 2016 |Volume 36 | Issue 4 | Yang XM et al. / Experimental & Information Study

Cool-related herbs on the cerebellum and stomach brum, and the small intestine. Compared with other were significant. Additionally, the effects of Warm- and herbs in Wuwei, in addition to the significant influ- Heat-related herbs on the middle spinal cord were sig- ence on the upper spinal cord, spinal cord, stomach, nificant. and medulla, the astringency herbs had a significant ef- When comparing Cold- and Cool-related herbs, the fect on the medulla. Cold-related herb had obvious effects on the large in- Compared with other herbs in Wuwei, the odorless testine and the lower spinal cord. The Cool-related herbs had a distinct, broad role on the kidney, spinal herbs had obvious effects on the heart and kidneys. cord, the upper spinal cord, and the lung. They both had effects on the cerebella and the stomach. Cumulative effects of Guijing The Warm-related herbs had obvious effects on the liv- The artificial neural network mined results between er, heart, and upper spinal cord, but the Heat-related Guijing and organs and tissues. herbs had obvious effects on the stomach, kidney, and Guijing significantly affected almost all of the nervous small intestine. They both had effects on the middle system: the brain, the cerebellum, the medulla, and the spinal cord (Table 4). spinal cord, which could be the biological basis for Gui- jing's role in affecting organ function, and was consis- Analysis results showed a cumulative effect with tent with our previous.17-19 Wuwei Gui Zang significantly affected the stomach. In addi- The artificial neural network mined the effects that tion to San Jiao Jing, Gui Fu had obvious effects on Wuwei had on organs and tissues. the large intestine, small intestine, and bladder, which Experimental results of association between Wuwei might be morphological basis of "transfer conversion and tissues and organs are shown in Table 5. Com- without preservation." Compared with other herbs in pared with other herbs in Wuwei, the main influence Guijing, San Jiao Jing and Xin Bao Jing had remark- of pungent and bitter herbs was on the stomach. The able effects on the medulla. Compared with Xin Bao pungent herbs had a significant effect on the stomach, Jing, San Jiao Jing had significant effects on the thy- as well as a significant effect on the small intestine, thymus, the brain, the upper spinal, and the medulla, mus, and middle spinal cord. The bitter herbs had a which might be one of biological reasons for "there are significant effect on the stomach, as well as a signifi- inner parts of intangible" for San Jiao Jing. Additional- cant effect on the heart, lung, and lower spinal cord. ly, Xin Bao Jing had significant effects on the stom- Compared with other herbs in Wuwei, the effects of ach, the kidney, the small intestine, and the medulla. salty and sweet herbs on the kidney were significant. (Table 6). The sweet herbs had a significant effect on the large intestine, the lower spinal cord, and cerebellum, in addi- DISCUSSION tion to the kidney. The salty herbs had a significant effect on the kidney, as well as a distinct effect on the The prediction of the organ and tissue reactions caused stomach, spinal cord, and the upper spinal cord. by Siqi, Wuwei, and Guijing consisted of a multi-vari- Compared with other herbs in Wuwei, the sour herbs able, non-linear problem. It was difficult to establish a played influential role on the stomach, heart, cere- precise and perfect non-linear system for a prediction

Table 6 Connection of Guijing and organs and tissues Guijing Related organs and tissues (score) Pi Jing Brain (41.12) Stomach (32.26) Small intestine (20.71) Cerebellum (15.14) Fei Jing Upper spinal (12.43) Stomach (10.79) Small intestine (10.3) Lung (9.61) Shen Jing Stomach (41.67) Cerebella (30.01) Brain (21.87) Lung (15.86) Gan Jing Heart (38.09) Stomach (33.66) Cerebellum (23.9) Lung (15.23) Xin Jing Stomach (29.44) Upper spinal (19.21) Thymus (15.21) Brain (7.24) Wei Jing Heart (22.17) Small intestine (18.26) Large intestine (17.13) Lung (10.49) Dachang Jing Small intestine (18.26) Large intestine (17.13) Lower spinal (9.81) Stomach (7.91) Pangguang Jing Lower spinal (18.47) Brain (16.94) Kidney (15.21) Cerebella (11.07) Dan Jing Kidney (19.08) Middle spinal (15.57) Lower spinal (12.51) Upper spinal (11.6) Xiaochang Jing Lung (13.26) Lower spinal (9.18) Brain (9.91) Small intestine (1.97) Sanjiao Jing Medulla (27.09) Thymus (24.35) Brain (31.2) Upper spinal (0.56) Xinbao Jing Medulla (13.95) Stomach (7.58) Kidney (6.59) Small intestine (4.3)

JTCM | www. journaltcm. com 544 August 15, 2016 |Volume 36 | Issue 4 | Yang XM et al. / Experimental & Information Study model using traditional methods.4 In recent years, rather than linear. Future studies should focus on col- through the development of computer technology, arti- lecting more information that could be applied to in- ficial intelligence technology, and the development of formation integration with regard to TCM. nonlinear science, studies focused on prediction models for multi-variable and non-linear problems have become a hot topic. Researchers have used the fractal, REFERENCES chaos, artificial neural network, and wavelet theories in nonlinear science; these theories have played a role in 1 Yu ZY, Wang B, Lu M. The preliminary study of Tradi- many fields, especially in the use of artificial neural net- tional Chinese Medicine four properities impacted on the works to establish prediction models with a good pre- of organs and tissue function. Shanghai Zhong Yi Za Zhi 2006; 40(4): 1-3. diction effect.15 Moreover, because of the characteristic extraction strategy usually combined with the human 2 Song LR. Chinese herbal medicine. Shanghai: Shanghai experience, it was possible to obtain a clear boundary Science and Technology Publishing House, 1999: 51-89. and state using imperfect information. 3 National Pharmacopoeia Commission. Pharmacopoeia of The artificial neural network is a computational mod- the people's Republic of China. Beijing: Chemical Indus- el. A multilayer feed-forward network that utilizes a BP try Publishing House, 2005: 25-125. neural network algorithm is the most widely used mod- 4 Qi MY. After using herbs many times, related relationship research among Siqi, Wuwei, Guijing and organs and tis- el that can approximate any non-linear curve and has a sues. Shanghai: Shanghai University of Traditional Chi- very strong non-linear mapping ability. The number of nese Medicine, 2011: 5-26. middle layers, the network processing unit for each lay- 5 Xu LY. Based source selection and extraction of the dy- er, and the network learning parameters can be set ac- namic characteristics of the sequence-level information fu- cording to a specific situation, which provides great sion model and algorithm. Shenyang: Northeastern Uni- flexibility. Therefore, it plays an important role in versity, 2001: 1-11. many application fields.16,17 6 Xu DC. The problem discussion of parameter estimation Each type of Siqi, Wuwei, and Guijing herbal effect is on the least square method. Chang Chun Shi Fan Da Xue associated with many organs and tissues. Analysis re- Xue Bao 2009; 28(2): 21-23. sults at different time points were not the same.16-19 Pre- 7 Li DP. The medical information analysis. Beijing: Tsing- sumably, TCM herbs affect organ and tissue functions hua University Publishing House, 2009: 160-162. in the first neuroendocrine organs and tissues, and the 8 Sun ZQ. Medical statistics. Beijing: People's Medical Pub- effect gradually extended, influencing other organs and lishing House, 2005: 22-23. tissues. To understand the impact position of tissue 9 Zhang YS. Artificial intelligence principle and applica- and organ functions under the cumulative action mode tion. Beijing: Higher Education Publishing House, 2004: of TCM, and to establish the relationship among herb 16-19. properties of Siqi, Wuwei, and Guijing and above in- 10 Gu LX, Shen JH. Artificial intelligent system guide. Bei- volving position, this study covered all herb properties jing: Mechanical Industry Publishing House, 2007: 25-32. of TCM and each property appeared at least five times. 11 Gong ZL, Chen Y, Su Y, et al. Application of data mining We measured excitement or inhibition of organs and in biomedical data analysis. Shanghai Jiao Tong Da Xue tissues, as reflected by uptake of 14C-2-deoxy glucose at Xue Bao 2010; 29(11): 1420-1423. 4 h after a single dose and repeated doses for a week of 12 Bao HF. The theory of biomedical knowledge integra- 39 herbs. Results showed indirect roles of TCM herbs tion. Zhejiang: Zhejiang University Publishing House, on each subsystem (organs and tissues) in the entire 2010: 200. body. By using the least-squares method to calculate 13 Zhang LM. Artificial neural network model and its ap- the effects of changed slope during treatment and the plication. Shanghai: Fudan University Press, 1993: 39-35. technology of an artificial neural network, we were able 14 Yang XM, Xu LY, Zhong F, et al. Data mining-based de- to extract the feature information that was connected tection of acupuncture treatment on juvenile myopia. to the cumulative changes in organ and tissue function Zhong Yi Za Zhi 2012; 32(3): 372-376. as a result of the herbal properties of Siqi, Wuwei, and 15 Yang LN. Borer, leaf miner occurrence prediction model Guijing. According to the influence values (scores) of research and application. Sichuan: University of Electronic different herb properties on different organs and tis- Science and Technology, 2009: 45-52. sues, we could draw the following conclusions: Deter- 16 Shi HF, Song J, Yang XM. Zhichan decoction indueces mining the correlation between the concept of Siqi, differentiation of dopaminergic neurons in Parkinson's dis- Wuwei, Guijing and the effect of Siqi, Wuwei, Guijing ease rats after neural stem cell transplantation. Zhong Guo on tissues and organs could not use a single organ or Shen Jing Zai Shen Za Zhi 2014; 9(9): 931-936. tissue, but rather required more than two kinds of or- 17 Yan MC, Ye FY, Zhang YQ, et al. Optimization model re- gan and tissue reactions. The effect of Siqi, Wuwei, search on efficacy in treatmentof chronic urticaeia by Chi- and Guijing on organs and tissues were non-linear, nese and Western Medicine based on a genetic algorithm.

JTCM | www. journaltcm. com 545 August 15, 2016 |Volume 36 | Issue 4 | Yang XM et al. / Experimental & Information Study

Zhong Yi Za Zhi 2013; 33(1): 60-64. 51-53. 18 Wang B, Yu ZY, Lu M. Guijing Morphological study 19 Yu ZY, Wang B, Lu M. Guijing Morphological study (2). ( 一). Shanghai Zhong Yi Yao Za Zhi 2006; 20(1): Shanghai Zhong Yi Yao Za Zhi 2006; 20(3): 32-36.

JTCM | www. journaltcm. com 546 August 15, 2016 |Volume 36 | Issue 4 |