Location: 100md.com > Paper > 2 > Text
.
NEURAI NETWORKS IN Q.S.A.R STUDIES:Estimation and prediction of The anticonvuisant activities of thirty-eight cinnamamides
http://www.100md.com    数理医药学杂志
HU Bu-chao WANG Hong-Lun GAO Guang-Jiang LI Xu-wu 数理医药学杂志 1999 0 12 3
关键词： 期刊 slyyxzz 0 译文园地 fur -->

(departmentof Nature product: Institute of Medicine Industry of shanxi XIAN710032)

Abstract Neural Networks(NN) Method was used to study The structure-activity relationship(S.A.R) of cinnamamide derivatives. The relationship between biological activity(P.C) and thosc parameters such as the partition coefficients(log p octanol/water)of the compounds Hammett δ constants and steric parameter(M.R) of cinnamamides were investigated by the modified backpropagation (MBP)Neural Netwoks. The biological activity of cinnamamides derivatives thus estimated andpredicted 100% fit with MLP Method. The resalts obtined by the developed MBPNN Method Seemto be better than those by Multivariate linear regression(MLR). The neural Networks Methodmight therefore be regarded as an excellent and effective chemometric Modelling techniquefor estim ating and predicting biological cativity on basic Q.S.A.R studies.
Key words: Neural Networks(NN) Cinnamamider Q.S.A.R

1 theory

1.1 principle of the Modified back-propagation(MBP) [ 1]

Fig1 Structure of neurons

Fig2 Multillayer neural networks

1.2 Algorithm
In this papper, Uses M.B.P algorith SupervisedLearning. Leaning Signal Consists of feed-forward and backpropagation(Fig3).

Fig3 Flow diaqram for MBP alqorithms

Alqorithm:
Ⅰ Sigmoid function
Ⅱ Connection Wij and Qi
Ⅲ Input traiming patten Sample Xvj and dvj
Ⅳ Calclate (coant) reality Value of oatput Yvj
Ⅴ regulate Weight Wij Wij (t+1)=Wij (t)+ηδvj Oi
j-out put layer: δj =yj (1-yj )(dj -yj )
j-hidden layer:
k is j(n+1) layer node.
Ⅵ Back to 4 Step Continued iteration untilerror of output becomes small enouh.

2 Data Sets

2.1 Programmingof Q.S.A.R. by NN
Accrding to characteristics of QuantitativeStructure-activity relationships(Q.S.A.R) and design of drug. We already Worked out “estimation and Prediction of the anticonvulsant acticvities ofthirtyeight cinnamamider Derivatires” Software.
The Fortran c Lanquag Program Worked Well on IBMPC486.
2.2 Prediction of activity
Data sets structure formula (I) of 38cinnamamide Derivative, Sample[2] .
Structure (I) X.R such and halogen, alkoxysabstituent etc.
Various Substituents give different activety (P.C)[3,4] .
According to group and X.R different we select38 Sample of known activieity join laning; Substituents X1 (logp) X2(∑) X3 (MR) ere selecte as of input signals (Input).
The degree of activity ED50 transformed Tolog1/c as output signal (output).
BY compareng The value calculated by NN, WithMLR The 1△lerror is minimal. The prediction of NN,excels that of MLR. shows NN. is an ideae method.

3 Results and Discussion

3.1 Network Parameters and Learing algorithm
Target ofConsecutive physco-active in Progress adopt NN(3-5-1) Network structures. found by followLearning with.
NN have bettereffective learning and convergence Speed(Fig4).

Fig4 Artificial Neural Network Specifications and Parameters

Parameter Simulated Data
Input nodes 3
Hidden nodes 5
Output nodes 1
Learning rate(ξ) 0.7
Momentam(η) 0.5
Gain 1
Transfer funetion Sigmoid
No of iteration 18000
Results of learning and training of 18,000 Time seeTab 1.
The nevral network have nonlinear handle andability of anti-noise, the result we calculate (estimatcd val ue) fit well with theexperiment result. (determine value) than the MLR mathod, the degree of precision is muchhigher order of magnitude is much improved.
3.2 Comparison of NN and MLR
Use logp partition coefficicnts (octanol/water) ∑δ(Hammts)δ (constants)MR(steric Parameter) as independent variable. and biological active log1/c as dependentvariable.
Using multivariate linear regression (MLR)architecture we built a Hansch eguation, which predicting.
log1/c=-0.155(±0.17)(logp)2 +1.350(±1.09)logp+0.257(±0.23)∑σ-0.264(±0.19)MR-2.032(±1.81)
(n=389,r=0.844,S=0.156,F=18.64,logp=4.35)
Tab1 It show that the Nearal netwark value weremuch more precislon than that of the Multivariate linear regression(MLR); Except sampleNO.14.22 the errors order of magnitude of MLR aremuch large Than that of N.N.
3.3 Conclasion
The study of prediction the Q.S.A.R of 38cinnamamides Derivative shows that:
Compared the new method with the linearmuliregression analysis in various way it was found that the nevral Network can be apotential tool in the routine Work of Q.S.A.R analysis.
It was shown that the neural Network can exceedthe level of the linear Multiregression analysis Thus we found that the neural networkmakes best use of the information included in the give data, resulting in an excellentgrading compared to other conventional mothods, moreover, the predictim ability in addtionto the easy operation indicated that the neural Network will being valuable tool indeveloping new drugs.

4 Acknowledgements

The authors are thankful to The science andTechnological commission of shaaxl province for providing financial assistance for theresearch project.

Tab.1Structural parameters and Biological activity of the cinnamamide Derivative (Substituted)

Sample X(x1 ) LOgp
X1 		∑σ
X2 		MR
X3 		Log 1/C
Est by MLR Est by NN
Obser Calcul 1Errorl obser calcul 1Errorl
01 3-Cl 3.43 0.37 0.80 0.788 0.658 0.130 0.788 0.790678 0.002678
02 3-F 2.86 0.34 0.29 0.578 0.572 0.006 0.578 0.583494 0.005494
03 4-F 2.86 0.06 0.29 0.458 0.499 0.041 0.458 0.452340 0.005660
04 4-Br 3.57 0.23 1.09 0.314 0.582 0.268 0.314 0.313917 8.270144e-05
05 2,4-Cl2 4.14 0.46 1.30 0.664 0.673 0.009 0.664 0.655334 0.008666
06 3,4-Cl2 4.14 0.60 1.30 0.550 0.709 0.159 0.550 0.546724 0.003276
07 4-Cl 3.43 0.23 0.80 0.606 0.621 0.015 0.606 0.602806 0.003194
08 3,4-OCH2 O- 2.66 -0.32 1.00 0.564 0.116 0.448 0.564 0.563792 0.000208
09 3,4,5-(OCH3 )3 2.66 0.07+ 1.68 0.793 0.037 0.756 0.793 0.792253 0.000747
10 4-NO2 2.44 0.78 0.94 0.268 0.292 0.024 0.268 0.290027 0.022027
11 3-NO2 2.44 0.71 0.94 0.324 0.274 0.050 0.324 0.301025 0.022975
12 3-CF3 3.60 0.43 0.70 0.921 0.744 0.177 0.921 0.902468 0.018532
13 2-CF3 3.60 0.54 0.70 0.723 0.772 0.049 0.723 0.723546 0.000546
14 4-CF3 3.60 0.54 0.70 0.921 0.772 0.149 0.921 0.723546 0.197454
15 3-OH,4-OCH3 2.05 -0.15 1.17 -0.272 -0.263 0.009 -0.272 -0.248678 0.023322
16 4-OCF3 2.70 -0.27 0.99 0.218 0.152 0.066 0.218 0.218390 0.000390
17 3-I 3.84 0.35 1.60 0.320 0.525 0.205 0.320 0.321269 0.001269
18 4-OC2 H5 3.19 -0.24 1.45 0.500 0.252 0.248 0.500 0.499724 0.000276
19 4-OC3 -7-n 3.77 -0.25 1.91 0.290 0.285 0.005 0.290 0.289360 0.00064
20 4-OC4 H9 -n 4.27 -0.32 2.37 0.180 0.196 0.016 0.180 0.179768 0.000232
21 3-Cl 3.67 0.37 0.80 0.410 0.717 0.307 0.410 0.413920 0.00392
22 3-F 3.10 0.34 0.29 0.495 0.674 0.179 0.495 0.301576 0.193424
23 4-F 3.10 0.06 0.29 0.495 0.602 0.107 0.495 0.500159 0.005159
24 4-Br 3.82 0.23 1.09 0.540 0.633 0.093 0.540 0.531801 0.008199
25 2,4-Cl2 4.38 0.46 1.30 0.735 0.680 0.055 0.735 0.738718 0.003718
26 3,4-Cl2 4.38 0.60 1.30 0.977 0.716 0.211 0.977 0.972233 0.004767
27 4-Cl 3.67 0.23 0.80 0.714 0.681 0.033 0.714 0.708546 0.005454
28 4-CF3 3.84 0.54 0.70 0.772 0.819 0.047 0.772 0.791499 0.019499
29 3-CF3 3.84 0.43 0.70 0.989 0.790 0.199 0.989 0.963252 0.025748
30 3-Cl 3.33 0.37 0.80 0.620 0.628 0.008 0.620 0.619791 0.000209
31 3-F 3.10 0.34 0.29 0.301 0.674 0.373 0.301 0.301576 0.000576
32 4-F 2.76 0.06 0.29 0.288 0.452 0.164 0.288 0.288056 5.626678e-05
33 4-Br 3.48 0.23 1.09 0.580 0.559 0.021 0.580 0.582673 0.002673
34 2,4-Cl2 4.04 0.46 1.30 0.600 0.665 0.065 0.600 0.609233 0.009233
35 4-Cl 3.33 0.23 0.80 0.801 0.592 0.209 0.801 0.798743 0.002257
36 3,4-Cl2 4.04 0.60 1.30 0.498 0.701 0.203 0.498 0.497898 0.000102
37 4-CF3 3.50 0.54 0.70 0.899 0.747 0.152 0.899 0.864690 0.03431
38 3-CF3 3.50 0.43 0.70 0.924 0.719 0.205 0.924 0.909313 0.014687
△ 国家自然科学基金项目

references

1 Zhen Yuzhao etal. Introduction to fuzzy theory and Neural Networks and Their Appliation.
2 Li Renli etal. Quantitative structure-Anticonvalsant activity relationships of cinnamamides actapharmaceutica sinical. 1986,21(8):580～585.
3 hansch C. LeoA. Substituent constants for correlation analysis in chemstry and biology. New YorkJohn wiler & sons.1979,49～52.
4 hansch. c.Quantitative Strcture-activety relationship in drug desing in Aricus Ej cd Drag Desiqn Voll: chapter 2. New York and Landon: Academic Press, 1971,300.

收稿日期：1998-09-21 (HU Bu-chao WANG Hong-Lun GAO Guang-Jiang LI Xu-wu)

 
.