йѧ
CHINA ENVIRONMENTAL SCIENCE
1999 19 6 Vol.19 No.6 1999



ȴ֬ЧϵŴģʽ
ƹա˱
ժҪְͨҵоѧʶЧӦ¼(RTECS)1998ϵͳ,Ѽ65ȴ֬ԴԶԿڷLD50,Ŵ罨ȴ֬ӽṹԴԶԹϵBPģʽ.Ԥ8ȴ֬Խ,нϺõʵ.ŴŴ㷨϶.ŴĹԼŴ罨Чϵ(QSAR)ģʽ.
ؼʣŴ㷨BP磻ȴ֬Чϵ
ͼţR994.6 ױʶ룺A
QSAR of chlorated aliphatic hydrocarbons by using GA-ANN approach.
TANG Gui-gang,BAI Nai-bing
(Research Center for Eco-Environmental Sciences,Chinese Academy of Sciences,Beijing 100085,China). China Environmental Science. 1999,19(6)539543
AbstractAcute toxicity data of 65 chorated aliphatic hydrocarbons to rat LD50 were collected from optical disk system of Registry of Toxic Effects of Chemical Substances (RTECS) developed by National Institute for Occupational Safety and Health (NIOSH). A model of artificial neural networks based on BP algorithm for relationships between acute toxicity of chorated aliphatic hydrocarbons to rat LD50 and their molecular structure descriptors got from structural formula were then built by genetic neural network approach. The mode was used to predict the toxicity of 8 chorated aliphatic hydrocarbons and the results showed that it worked well. Genetic neural network was obtained by integrating genetic algorithm and BP neural networks. How a genetic neural network was constructed and how a QSAR model was built by using it was described in detail in this paper.
Key wordsgenetic algorithmBP neural networkschloro-aliphatic hydrocarbonsQSAR
ճ,ǾӴжȴ֬໯,Ȼˮкжĵͼȴ֬Ⱦ,ֱΣǵĽ.,ȴ֬ĶоҪ.ȴ֬෱,൱δжԲ,ּԤķӽṹ뻷ЧӦϵоʵʼֵ[1-3].Ŵ㷨(GA)˹(ANN)õŴ(Genetic Neural Networks,GNN)ӦԽģ,еȹģϵ.ͼӦŴоȴ֬˽ṹ䶾ԵĹϵ.
1 ȴ֬Զ
ȴ֬ԴСļԶȡְҵоѧʶЧӦ¼(RTECS)1998ϵͳ,ʹõǴڷLD50(mg/kg),ѧϰԤʱȡ,ѧϰ65,ݼ1.
2 ȴ֬Թϵ˹ģʽ[4,5]
2.1 ˹ģʽĻ
ûڷ򴫲㷨(Backpropagation Algorithm,BP)Ķǰ.Ԥȴ֬ʱ,Ϊ˻ȡһʵõBPģʽ,Ҫȷ²,ȴ֬ӽṹ;˽ṹѵ,BP˽ṹڵ,ѵѧϰa h ڵĳʼȨֵ.Щ˼·,,ȴ֬ƳǶӰķӽṹ;ٸܹȡԤЧԭ,Ŵ㷨BPϻŴ,ɸѡ뽨ģ,ͬʱŻBP˽ṹѵ,ʵõģʽ.
1 ȴ֬˽ṹԴԶԿڷLD50(mg/kg)
ϵģʽ*
Table 1 The relationships between molecular topological structure descriptors and acute toxicities of chloro-aliphatic hydrocarbons to rat LD50 (mg/kg)
ƽṹԶ
LOG(ڷLD50)
ABCDEFGHIJKLʵֵֵ
ȼ1310010000003.2553.261
ȼ1220011000003.2043.210
ȼ1130010100002.9582.957
ȼ1040010000003.3713.378
1,1-2420011000002.8602.872
1,2-2420020000002.8262.825
1,1,1-2330010100003.9823.985
1,1,2-2330021000002.9222.927
1,1,1,2-2240020100002.8262.825
1,1,2,2-2240022000002.3982.411
2150021100002.9632.967
2060020200003.6493.651
1-ȱ3710010000003.3013.314
1,1-ȱ3620011000003.8133.794
1,1,1-ȱ3530010100003.8733.888
1,1,2-ȱ3530021000003.0903.070
1,2,2-ȱ3530021000003.0573.070
1,2,3-ȱ3530030000002.7032.697
1,1,1,3-ȱ3440020100002.6902.681
1,1,2,2,3-ȱ3350032000002.9132.901
1-ȶ4910010000003.4273.840
2-ȶ4910010000004.2423.840
1,1-ȶ4820011000003.6873.734
1,2,3,4-ȶ4640040000002.3522.352
1,1,2,3,4,4-ȶ4460042000002.6022.597
1,1,1,5-5840020100002.6332.637
1,1,1,7-ȸ71240020100002.6782.686
1,1,7,7-ȸ71240022000003.3983.403
1--5-׻81710010000003.8663.863
1,1,1,9-91640020100002.5682.568
ȹ102110000000004.6564.653
Ȼ3150032000001.8921.891
1,2,3,4,5,6-Ȼ6660060000002.0001.997
-Ȼ6660160000002.2482.245
-Ȼ6660260000003.7783.778
-Ȼ6660360000001.8811.882
-Ȼ6660460000003.0003.000
ϩ2311010010002.6992.696
1,1-ϩ2221011001002.3012.302
1,2-ϩ2221020020002.8862.885
ϩ2131021011003.7523.752
ϩ2041022002003.4203.434
1-ȱϩ3511010010003.2903.288
3-ȱϩ3511010000012.6632.660
1,3-ȱϩ3421020010012.6722.579
1,2-ȱϩ3421020020003.3013.287
2,3-ȱϩ3421020010012.5052.573
1,2,3-ȱϩ3331030020012.7902.768
1,1,2,3-ȱϩ3241031011012.5442.524
1--2-׻ϩ4711010010003.6503.648
1,4--2-ϩ4621020000021.9491.946
3--2(ȼ׻)-ϩ4621010000012.1762.172
1,2--2-׻ϩ4621020010002.1762.173
2,3,4--2-ϩ4531030020012.5322.530
1,2,4--2-ϩ4531030010022.3502.349
1,1,2,4--2-ϩ4441031010012.6252.621
1,3,4,5,6-Ȼϩ6551050000103.4673.468
2--1,3-ϩ4512010010012.6532.655
1,1,2--1,3-ϩ4332021011002.8322.837
1,1,2,3--1,3-ϩ4242031021022.6242.625
1,1,2,3,4,4--1,3-ϩ4062042022021.9141.913
ȫȻϩ5062051000003.1143.112
2,3,4,5-ȼϩ6443040040022.5682.568
3--3-׻-1-Ȳ6910010000013.1763.176
-4-Ǽ׻ϩ6081053000003.5683.568

ע:* е2.2ĶӦϵA-[1]B-[2]C-[3]D-[4]E-[11]F-[12]G-[13]H-[16]I-[18]J-[19]K-[21]L-[23]
2.2 ȴ֬ӽṹ
ݷӵ23ֿܶȴ֬Բ(ֻ)ķӽṹ:[1]̼ԭ [2]ԭ [3]ԭ [4]ϩ [5]ϩ [6]ϩ [7]Ȳ [8]ṹ [9]ʽṹ [10]˳ʽṹ [11]a b g d ṹ[12]ClԭӵCԭ[13]һClԭӵCԭ [14]ClԭӵCԭ[15]4ClԭӵCԭ,[16] [23]Ϊӽṹ:

2.3 Ŵ
ȷ˽ṹѵͳ,ѵķӽṹͨҲǿʹ(ɷַ)ɸѡ,ͨ籾.õģʽһкõԤ,ʱϡ.Ŵ㷨BPϻõŴ,ŻBP˽ṹѧϰ,ɸѡ뽨ģķӽṹ.Ŵ㷨һ㷨,Դڴĵ˼,ͨŴӡ(Replication)ӽ(Crossover)ͱ(Mutation)ȺŴ,ʤ̭ɸѡĸ.ͨŴ㷨,,ӶнϺԤЧģʽ.Ŵ㷨4.
2.3.1 ʼŴ㷨ȺеĸŴ,Ŵ㷨ŴöƱ,еBP2.ŴӵĽṹ2,ɱ2֪,Ŵӹ41λ,Ŵ㷨Ҫ241ĽռѰŽӽŽ.
2 2ŴŴӽṹ
Table 2 The structure of gene of genetic neural networks with two hidden layers
λ 
0Ƿеһ,1-,0-
14һڵ(116)
5Ƿеڶ,1-,0-
69ڶڵ(116)
1012ѧϰ{0.9,0.8,0.7,0.6,0.5, 0.25,0.125, 0.0615} 
1315{0.9,0.8,0.7,0.6,0.5,0.4,0.3,0.2}
1617ʼȨֵ{+1.0,+0.5,+0.25,+0.125}
1840ṹ,ÿһλӦһ,1-òѧϰ,0-òѧϰ

2.3.2 Ŵӽ,õ˽ṹѵԼѧϰķӽṹѵ,Ӧֵ.ΪʹнϺõѧϰнϺõԤ,ڼŴ㷨ĸӦֵʱ,ѡȡһ(90%)ѧϰѵ,ѵõʣµԤ,˷ֱÿԤһ.ѵԤıȨƽ,þֵĵΪӦֵ.
2.3.3 Ŵ(ơӽͻ)Ŵ㷨Ⱥ,һȺ.
2.3.4 Ҫ,㷨,,ת2.3.2.
3
Ա2,ֱǵ23ֽṹ,Ŵɸѡ.C++п,PII΢ʵ.Ŵ㷨ҪȺģ50,ӽ0.6,0.005.Լ1500(30)Ժ,󲿷ָ3ʾģʽ.øģʽͱ2ѵ,3,ϵr=0.992.
3 ȴ֬˽ṹڷLD50ϵģʽṹ
Table 3 The structure of BP neural network for rat LD50 of chloro-aliphatic hydrocarbons
 Ŀ 
μӽģĽṹ2.2г[1][2][3][4][11][12][13][16][18][19][21][23](12)
ѧϰa 0.8
h 0.6
ڵ2,ڵֱΪ155
ڵʼȨֵ0.5

Ϊ˼ѵõģʽԤ,һԤ,4ĴڷLD50֪,4ڷLD50ֵδ֪,ǵСڷLD50ֵ֪.ΪСڷLD50ûͬ,ֵӦͬһ,СڷLD50ԸڷLD50Ԥֵ׼ȷ.4г8ݵԤ.
4ԤԤ
Table 4 Data of chloro-aliphatic hydrocarbons used to test the BP neural networks
ĿṹԶLOG(ڷLD50)
ABCDEFGHIJKLʵֵֵ
1,2-ȱ81710010000003.292.99
1--2-׻2221020020003.873.86
ʽ-1,2-ϩ4351041020013.092.89
1,1,2,3,4--2-ϩ4621020000012.672.90
3,4--1-ϩ4621020010012.85*2.05
ʽ-1,3--1-ϩ4441040020022.58*2.38
1,2,3,4--2-ϩ4351041011011.46*2.46
1,1,2,3,4--2-ϩ2330021000002.94*3.12

ע:*СڷLD50,ԤĴڷLD50շӽṹALĺͬ1
ɱ4֪,֪ڷLD50ԤֵʵֵȽϽӽ,δ֪ڷLD50ԤֵϺ͸СڷLD50ͬһ,ֻ1,2,3,4--2-ϩСڷLD50͸ڷLD50Ԥֵϴ,ԤֵҲڸ߶ֵ[6],һĲοֵ.
һΪ,ȴ֬ĶԹԭ[1-3].ȷ½ṹ,ɾ綾໯.ϩӽṹķӶԽϸ,ϩp ԭӵp ,ԭӵķӦܼǿ,ϩϩҲƵĶ.ԭ,ӷӦ,ʹӶ.̼ԭӶ,󽫲ڶĿӽӴ뷴Ӧ,.ŴɸѡڽģĽṹ̼ԭԭһЩӽṹ,Ȼǵʶһ.,ŴɸѡĽṹԭԭӵ̼ԭԭӵ̼ԭ,Щ֬еķֲй,ӰӵĻ.
4 
ŴڽQSARģʽͬʱʵѡŻ,ȴ֬ӽṹԴԶԹϵBPģʽнϺõʵ.
ĿȻѧĿ(29877032)
߼飺ƹ(1973-),,ˮ,йѧԺ̬о˶ʿ.Ҫ»ȾﹹЧϵо.?/FONT>ˮ滮?/FONT>о,ڽйȻѧĿQSAR˹о?/FONT>о.2ƪ.
ߵλйѧԺ̬о, 100085
ο
[1]Rorjie Enniel, Eriksson Lennart, Verboom Hans, et al . Predicting reductive transformation rates of halogenated alphatic compounds using different QSAR approaches [J]. Environ. Sci. Pollu. Res. Int., 1997,4(1):47- 54.
[2]Eriksson Lennart, Sjoestroem Michael, Svante Wold. Rational ranking of chemicals according to environmental risk [J]. Chemom. Intell. Lab. Syst., 1992, (14):245- 252.
[3]Eriksson Lennart, Sjoestroem Michael, Berglind Rune, et al. Multirariate biological profiling of the subacute effects of halogenated aliphatic hydrocarbons [J]. J. Environ. Sci. Health, Part A, 1993, A28(5):1123- 1144.
[4]Pao Yoh-han. Adaptive pattern recognition and neural networks [M]. Addison-Wesley Publishing Company, Inc, 1989.117- 136.
[5] ,ɽ,ع. ֵ㷨? Ŵ㷨 [M]. :ѧ,1995.137- 154.
[6]ҽѧٿȫд. йҽѧٿȫ? ѧ [M]. Ϻѧ,1982.47.
ոڣ1999-04-06
