Quantitative structure property relationship models for the prediction of liquid heat capacity

Quantitative structure property relationship models for the

prediction of liquid heat capacity
Xiaojun Yaoa, b, Botao Fana *, J. P. Douceta, A. Panayea, Mancang Liub, Ruisheng Zhangb, Xiaoyun Zhangb, Zhide Hub*
a
Universite Paris 7-Denis Diderot, ITODYS-CNRS UMR 7086, 1, Rue Guy de la Brosse, 75005 Paris, France
b
Department of Chemistry, Lanzhou University, Lanzhou 730000, China

Abstract

Quantitative Structure-Property Relationship (QSPR) neural networks (RBFNNs) were used to construct the
models based on molecular descriptors derived from QSPR models. The root mean square errors in liquid heat
molecular structures have been developed for the predic- capacity predictions for the training, test and overall data
tion of liquid heat capacity at 25 8C using a diverse set of sets are 16.857, 18.744 and 17.141 heat capacity units,
871 organic compounds. The molecular descriptors used to respectively. The prediction results are in agreement with
represent molecular structures include constitutional and the experimental values, but the RBFNN model seems to
topological indices and quantum chemical parameters. be better than stepwise regression method.
Forward stepwise regression and radial basis function

1 Introduction present, there exists a great number of molecular descriptors

that encode constitutional, topological, geometry and
The study of the quantitative relationship between prop- electronic features of organic compounds [3 ± 5]. Among
erty/activity and molecular structure (QSPR/QSAR) is an various structure descriptors, those derived from molecular
important research area in computational chemistry and has structure alone have a particular advantage of the possibility
been widely used in the prediction of physicochemical to calculate them based only on molecular structural feature
properties and biological activities of organic compounds and to be applicable to different families of compounds [6 ±
[1 ± 2]. This kind of study can not only develop a method for 7]. After the calculation of molecular descriptors, linear
the prediction of the property under investigation of new methods, such as multiple linear regression (MLR), princi-
compounds that have not been synthesized but also can pal component regression (PCR) and partial least squares
identify and describe important structural features of (PLS) or non-linear methods, such as many types of artificial
molecules that are relevant to variations in molecular neural networks, can be used in the development of a
properties, thus gain some insight into structural factors mathematical relationship between the structural descrip-
affecting molecular properties. To develop a QSPR model, tors and the property to be predicted.
the following steps are usually involved: data collection, Physical and thermodynamic property data of organic
molecular geometry optimization, molecular descriptors compounds such as liquid heat capacity are important in the
generation, descriptors selection, model development and engineering design and operation of industrial chemical
finally model performance evaluation. processes [8]. In liquid phase reactions, the liquid heat
One of the important problems in QSPR is the description capacity is often required to evaluate the energy variation
of molecular structures using molecular descriptors, which for the liquid chemical reactants involved in the reaction.
can include structural information as much as possible. At The heat capacity of a compound is defined as the amount of
energy needed to raise a unit of the substance in question by
one unit of temperature. The liquid heat capacity data is also
* To receive all correspondences
needed in the heat exchanger and energy balance design
[email protected] calculations. Since the experimental determination of heat
capacity is both time-consuming and expensive, and there is
Key words: Radial basis function networks; QSPR; Molecular increased need of reliable physical and thermodynamic data
descriptor; heat capacity for the optimization of chemical processes, it would be very
useful to develop predictive models that can be used to
Abbreviations: Quantitative Structure-Property Relationship predict these properties of organic compounds that are not
(QSPR); Radial basis function neural networks (RBFNNs) synthesized or their properties are unknown.

QSAR Comb. Sci. 22 (2003)  WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 1611-020X/03/0104-0029 $ 17.50+.50/0 29
 X. Yao et al.

Although many reported QSPR methods have been logical descriptors. Constitutional descriptors are basically
successfully used to predict a diverse set of physicochemical related to the number of atoms and bonds in each molecule.
properties, their use in predicting heat capacity is rather Topological descriptors include valence and non-valence
limited [9 ±10]. Gakh et al. obtained a QSPR model for heat molecular connectivity indices calculated from the hydrogen-
capacity of alkanes using graph theory descriptors and neural suppressed formula of the molecule, encoding information
networks with a RMS of 4.04 heat capacity unit [9]. Liu et al. about the size, composition and the degree of branching of a
developed several QSPR models of alkanes using their molecule. The quantum chemical descriptors include infor-
molecular electronic edge vectors (MEDV). The RMS for the mation about binding and formation energies, partial atom
heat capacity of 134 alkanes was 3.81 heat capacity unit [10]. charge, dipole moment, and molecular orbital energy levels.
But their works are limited to the investigations of a same After the calculation of molecular descriptors, forward
family of compounds (alkanes). In our previous works we stepwise regression method was then used to select the
have successfully developed several QSPR models based on significant descriptors to develop QSPR model of the liquid
RBFNNs for the prediction of physicochemical properties heat capacity.
[11 ±15] and HPLC retention indices of N-Benzylideneani-
lines [16]. The goal of the present study is to extend our
2.3 Radial basis function neural networks
previous investigations in order to, for the first time, establish
a QSPR model that can predict the liquid heat capacity at The theory of RBFNNs and its local modeling ability have
258C for a diverse set of organic compounds dependent only been well described in a recent paper by Walczak and
upon their molecular structures. MLR and RBFNNs are Massart [19]. RBFNNs can be described as a three-layer
applied to establish quantitative linear and non-linear feedforward structure. As presented schematically in Fig-
relationships between heat capacity and molecular descrip- ure 1, the RBFNNs consist of three layers: input layer,
tors respectively. The data set used in our work is more diverse hidden layer and output layer. The input layer does not
and the models developed are more general and practical process the information; it only distributes the input vectors
with respect to the models reported early by other authors. to the hidden layer. The hidden layer of RBFNNs consists of
a number of RBF units (nh) and bias (bk). Each hidden layer
unit represents a single radial basis function. Each neuron
on the hidden layer employs a radial basis function as non-
2 Materials and Methods
linear transfer function to operate on the input data. The
most often used RBF is a Gaussian function that is
2.1 Data set
characterized by a center position (cj) and width (rj). The
The data used in this investigation were collected from RBF functions by measuring the Euclidean distance be-
literature [8]. The set consists of 871 organic compounds. tween input vector (x) and the radial basis function center
The data set includes hydrocarbons, fluorocarbons, chlor- (cj) and performs the non-linear transformation with RBF in
ocarbons, bromocarbons, iodocarbons, alcohols, acids, ke- the hidden layer as given in below:
tones, aldehydes, ethers, esters, amines, nitriles, etc. A  
complete list of the compound names and their correspond- hj x† ˆ exp x cj 2 =r2j 1†
ing experimental liquid heat capacities at 25 8C is given in
Table 1. The data were randomly separated into two subsets:
a training set of 746 compounds and a test set of 125
compounds. The training set was used to adjust the
parameters of RBFNNs and the test set was used to evaluate
the predictive ability of RBFNNs.

2.2 Molecular descriptors generation and regression

analysis
The calculation of molecular descriptors is described as
below: all molecules were drawn into Hyperchem [17] and
pre-optimized using MM‡ molecular mechanics force field.
A more precise optimization is done with semi-empirical
PM3 method in Hyperchem and thereafter quantum
chemical descriptors were obtained. All calculations were
carried out at RHF level with no configuration interaction.
The molecular structures were optimized using the Polak-
Ribiere algorithm until the root mean square gradient was
0.01. The resulted geometry was then transferred into
software Dragon [18] to calculate constitutional and topo- Figure 1. The typical architecture of the RBFNNs

30 QSAR Comb. Sci. 22 (2003)

Quantitative structure property relationship models for the prediction of liquid heat capacity

Table 1. The compounds and the predicted results of liquid heat capacities: (J/mol ´ K)
Number NAME HEATCAP MLR RBFNNs
1 Bromochlorodifluoromethane 127.81 106.55 127.23
2* Bromotrichloromethane 130.65 108.16 130.03
3 Bromotrifluoromethane 143.25 104.50 126.00
4 dibromodifluoromethane 129.94 109.53 129.03
5 dichlorodifluoromethane 121.07 105.12 125.43
6 phosgene 112.03 118.18 117.02
7 trichlorofluoromethane 68.65 108.15 128.95
8 carbontetrachloride 130.72 100.65 126.01
9* tribromomethane 129.94 141.93 136.20
10 chlorodifluoromethane 111.66 100.40 105.96
11 dichlorofluoromethane 107.20 104.37 106.08
12 chloroform 112.49 97.81 99.01
13 bromochloromethane 100.75 113.78 100.37
14 dibromomethane 105.11 129.51 116.57
15 dichloromethane 101.98 95.29 82.05
16* difluoromethane 107.05 101.92 94.33
17 diiodomethane 135.45 145.06 141.58
18 formic acid 98.40 98.60 106.48
19 methylbromide 85.16 119.67 81.05
20 methylchloride 82.33 104.69 61.86
21 methyliodide 82.91 128.28 98.97
22 nitromethane 104.22 111.55 127.14
23* methanol 79.93 123.08 94.04
24 Methylmercaptan 96.39 131.43 92.57
25 Methylamine 114.19 120.29 99.43
26 Bromotrifluoroethylene 135.33 135.29 140.92
27 1,2-dibromotetrafluoroethane 184.03 153.13 179.99
28 Chlorotrifluoroethylene 150.36 135.29 140.92
29 Chloropentafluoroethane 174.61 150.71 173.07
30* 1,2-dichlorotetrafluoroethan 110.13 150.31 174.06
31 1,1,2-trichlorotrifluoroetha 169.44 153.97 177.66
32 Tetrachloroethylene 146.93 134.64 140.24
33 trichloroacetylchloride 168.81 161.40 169.97
34 2-chloro-1,1-difluoroethylen 126.95 140.04 138.28
35 trichloroethylene 123.67 130.81 128.04
36 dichloroacetyl chloride 154.23 152.81 152.72
37* trichloroacetaldehyde 153.22 150.51 154.35
38 pentachloroethane 190.91 146.28 163.41
39 trifluoroacetic acid 170.15 155.99 160.88
40 pentafluoroethane 168.85 143.59 157.87
41 1,1,2,2-tetrabromoethane 168.43 157.56 156.62
42 1,1-dichloroethylene 112.43 122.69 121.29
43 cis-1,2-dichloroethylene 117.88 116.74 110.47
44* trans-1,2-dichloroethylene 118.37 115.54 109.88
45 chloroacetyl chloride 142.92 138.03 137.11
46 dichloroacetaldehyde 134.07 141.23 138.66
47 dichloroacetic acid 172.72 164.08 161.64
48 1,1,1-trichlorofluoroethane 151.42 147.95 154.60
49 1,1,1,2-tetrachloroethane 154.63 144.61 151.58
50 1,1,2,2-tetrachloroethane 159.39 133.99 139.18
51* 1,1,1,2-tetrafluoroethane 141.80 135.99 145.90
52 vinyl bromide 103.13 109.82 106.96
53 vinyl chloride 86.01 106.30 101.32
54 1-chloro-1,1-difluoroethane 132.37 127.27 140.17
55 acetyl chloride 115.02 114.12 121.16
56 chloroacetaldehyde 115.19 124.99 124.32
57 methyl chloroformate 141.03 114.12 121.16
58* 1,1,1-trichloroethane 145.56 137.87 147.19
59 1,1,2-trichloroethane 151.46 131.14 130.57
60 1,1,1-trifluoroethane 139.86 122.51 137.18
61 acetonitrile 91.94 99.29 88.93
62 1,1-dibromoethane 133.61 127.47 128.10

QSAR Comb. Sci. 22 (2003) 31

 X. Yao et al.

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

63 1,2-dibromoethane 131.75 125.05 122.86
64 1,1-dichloroethane 131.00 128.86 128.18
65* 1,2-dichloroethane 128.77 112.69 109.94
66 1,1-difluoroethane 100.76 124.00 129.91
67 1,2-difluoroethane 112.25 128.48 120.97
68 acetaldehyde 102.72 108.19 113.01
69 acetic acid 128.66 118.59 132.33
70 methyl formate 113.87 131.35 133.01
71 bromoethane 100.54 109.17 107.08
72* ethyl chloride 106.00 113.32 109.88
73 2-chloroethanol 116.14 146.96 145.79
74 ethyl fluoride 117.84 123.42 115.86
75 ethyl iodide 111.06 122.57 121.57
76 ethanol 107.40 129.87 130.32
77 ethylene glycol 165.52 144.69 148.26
78 dimethyl sulfate 189.12 217.09 189.08
79* ethyl mercaptan 122.54 120.88 116.68
80 dimethyl disulfide 148.08 140.96 144.75
81 dimethylamine 147.08 131.36 130.96
82 ethylamine 117.47 130.38 135.88
83 monoethanolamine 193.21 147.22 154.69
84 ethylenediamine 169.25 146.61 157.06
85 hexafluoropropylene 176.12 176.98 188.68
86* hexafluoroacetone 225.95 179.79 195.67
87 octafluoropropane 225.61 195.53 209.05
88 acrylonitrile 108.79 113.63 110.76
89 2,3-dichloropropene 130.87 144.46 143.05
90 acrolein 121.07 123.43 128.40
91 acrylic acid 148.96 141.17 149.58
92 vinyl formate 141.93 141.17 149.58
93* 3-bromo-1-propene 127.35 120.48 122.16
94 2-chloropropene 120.57 123.85 126.25
95 3-chloropropene 112.83 122.23 123.24
96 a-epichlorohydrin 133.04 129.14 136.54
97 methyl chloroacetate 163.63 168.59 170.25
98 ethyl chloroformate 198.57 174.88 172.35
99 1,2,3-trichloropropane 172.40 153.64 150.64
100* 3-iodo-1-propene 139.62 129.91 134.05
101 propionitrile 120.92 130.64 126.06
102 hydracrylonitrile 201.51 167.10 161.71
103 propylene 115.30 106.57 109.28
104 1,2-dibromopropane 160.68 148.16 148.89
105 1,1-dichloropropane 153.16 150.93 149.69
106 1,2-dichloropropane 145.81 148.95 149.83
107* 1,3-dichloropropane 149.59 143.22 141.77
108 2,2-dichloropropane 126.88 137.19 148.45
109 1,2-diiodopropane 190.49 156.10 157.86
110 acetone 127.53 119.95 131.94
111 allyl alcohol 143.67 141.62 146.36
112 methyl vinyl ether 128.31 139.83 140.30
113 n-propionaldehyde 129.26 132.82 138.77
114* ethyl formate 149.98 156.18 159.08
115 methyl acetate 138.99 157.43 161.81
116 propionic acid 158.44 141.34 152.25
117 3-mercaptopropionic acid 202.00 172.61 175.82
118 methoxyacetic acid 190.82 173.60 176.48
119 1-bromopropane 139.24 130.80 132.96
120 2-bromopropane 142.24 125.92 129.48
121* isopropyl chloride 133.71 128.35 131.65
122 n-propyl chloride 129.29 132.45 132.09
123 1-fluoropropane 101.44 136.43 134.14
124 isopropyl iodide 136.98 128.16 133.19

32 QSAR Comb. Sci. 22 (2003)

Quantitative structure property relationship models for the prediction of liquid heat capacity

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

125 n-propyl iodide 135.08 145.17 147.50
126 allylamine 146.40 144.18 152.20
127 1-nitropropane 166.88 172.95 179.75
128* 2-nitropropane 171.00 171.36 177.32
129 propane 117.67 115.45 113.91
130 isopropanol 172.39 143.30 151.02
131 methyl ethyl ether 140.81 154.53 151.05
132 n-propanol 144.32 153.55 157.20
133 2-methoxyethanol 171.15 179.43 177.57
134 1,2-propylene glycol 212.32 164.98 169.75
135* 1,3-propylene glycol 243.40 170.70 174.44
136 glycerol 260.94 188.29 189.91
137 n-propyl mercaptan 145.40 143.04 142.98
138 isopropyl mercaptan 145.76 131.23 135.29
139 ethyl methyl sulfide 144.50 146.96 154.87
140 n-propylamine 168.38 156.11 163.28
141 isopropylamine 168.18 145.89 154.93
142* trimethylamine 136.78 149.43 150.05
143 1-amino-2-propanol 182.88 169.15 175.61
144 3-amino-1-propanol 204.23 174.63 180.30
145 1,2-propanediamine 203.98 171.01 178.79
146 decafluorobutane 261.96 245.17 246.30
147 furan 112.70 100.83 113.55
148 thiophene 124.15 114.08 125.44
149* methacrylonitrile 147.42 139.60 138.87
150 pyrrole 131.74 94.84 113.15
151 vinylacetonitrile 131.04 142.50 140.56
152 1,3-butadiene 122.26 132.94 105.84
153 1,3-dichloro-trans-2-butene 173.16 160.03 160.79
154 1,4-dichloro-cis-2-butene 175.23 155.76 156.20
155 1,4-dichloro-trans-2-butene 172.85 157.92 157.10
156* 3,4-dichloro-1-butene 165.48 161.19 159.77
157 2,5-dihydrofuran 127.86 130.55 136.87
158 divinyl ether 131.61 149.42 152.09
159 methacrylic acid 174.60 169.33 172.58
160 methyl acrylate 158.13 178.94 178.87
161 vinyl acetate 155.86 169.33 173.20
162 acetic anhydride 189.75 175.90 179.95
163* n-butyronitrile 155.72 153.45 150.10
164 isobutyronitrile 156.35 151.11 150.21
165 3-methoxypropionitrile 162.47 190.29 183.65
166 1-butene 130.74 131.61 136.01
167 cis-2-butene 126.25 124.72 128.85
168 trans-2-butene 130.89 124.52 128.85
169 isobutene 131.72 129.78 136.34
170* 1,2-dibromobutane 165.41 176.56 174.35
171 2,3-dibromobutane 163.65 168.77 169.58
172 1,4-dichlorobutane 184.25 172.48 172.94
173 1,4-diiodobutane 181.82 185.88 184.43
174 n-butyraldehyde 159.03 159.50 164.04
175 isobutyraldehyde 166.99 157.06 160.68
176 methyl ethyl ketone 159.75 151.62 159.38
177* ethyl vinyl ether 148.27 164.72 165.49
178 tetrahydrofuran 119.08 141.68 148.03
179 cis-2-butene-1,4-diol 252.75 187.92 188.44
180 isobutyric acid 174.56 178.16 182.62
181 n-butyric acid 182.09 180.46 186.68
182 ethyl acetate 170.66 184.67 188.55
183 methyl propionate 179.33 190.45 191.94
184* n-propyl formate 166.88 183.69 185.76
185 1-bromobutane 152.68 156.06 158.30
186 2-bromobutane 169.43 154.09 156.44

QSAR Comb. Sci. 22 (2003) 33

 X. Yao et al.

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

187 n-butyl chloride 158.13 156.83 157.55
188 sec-butyl chloride 158.13 156.10 156.91
189 tert-butyl chloride 159.39 142.56 153.41
190 2-iodo-2-methylpropane 139.08 138.92 150.63
191* N,N-dimethylacetamide 170.59 191.54 192.85
192 n-butane 140.84 139.60 141.63
193 isobutane 141.61 134.66 138.48
194 n-butanol 160.12 179.89 183.38
195 sec-butanol 172.18 173.70 178.16
196 diethyl ether 175.81 179.11 177.11
197 methyl-propyl-ether 164.40 178.69 177.11
198* methyl isopropyl ether 158.70 172.98 173.40
199 isobutanol 163.33 178.16 180.38
200 1,3-butanediol 236.42 192.37 195.69
201 1,4-butanediol 241.24 199.32 201.59
202 2,3-butanediol 221.65 188.04 192.19
203 1,2-dimethoxyethane 186.52 204.00 198.92
204 2-ethoxyethanol 211.06 205.98 204.49
205* diethylene glycol 258.37 224.76 221.77
206 n-butyl mercaptan 173.62 168.12 168.71
207 isobutyl mercaptan 170.67 165.69 166.46
208 sec-butyl mercaptan 171.89 159.14 160.83
209 tert-butyl mercaptan 175.08 146.16 155.47
210 diethyl sulfide 171.19 162.61 168.66
211 isopropyl methyl sulfide 163.31 173.08 178.50
212* methyl propyl sulfide 174.99 174.75 181.15
213 diethyl disulfide 206.15 180.28 182.58
214 n-butylamine 178.56 183.91 190.08
215 isobutylamine 183.12 182.49 186.43
216 sec-butylamine 171.78 177.24 181.98
217 tert-butylamine 194.09 163.86 173.11
218 diethylamine 170.40 180.20 184.32
219* dimethylethanolamine 221.16 206.00 204.75
220 2-aminoethoxyethanol 253.17 229.81 229.01
221 N-aminoethylethanolamine 270.10 228.98 231.04
222 furfural 169.03 162.87 159.06
223 pyridine 133.85 127.18 132.48
224 cyclopentadiene 118.30 101.69 117.46
225 2-methyl-1-butene-3-yne 135.31 136.67 139.80
226* 1-pentene-3-yne 142.59 142.18 142.41
227 1-pentene-4-yne 138.08 139.09 143.47
228 glutaronitrile 184.45 179.08 175.04
229 furfuryl alcohol 202.77 178.89 175.34
230 3-methylthiophene 144.22 154.43 155.23
231 N-methylpyrrole 150.61 134.49 142.98
232 ethyl cyanoacetate 230.24 228.21 221.08
233* cyclopentene 114.06 112.89 127.33
234 isoprene 144.75 144.73 148.09
235 2-methyl-1,3-butadiene 153.17 146.04 149.87
236 1,4-pentadiene 150.17 146.12 152.00
237 1-pentyne 163.56 149.13 152.17
238 2-pentyne 174.09 153.55 152.99
239 3-methyl-1-butyne 165.30 147.84 150.63
240* cyclopentanone 164.16 151.35 158.06
241 methyl isopropenyl ketone 158.17 170.24 174.89
242 acetylacetone 204.44 178.45 184.79
243 allyl acetate 191.62 201.33 202.42
244 ethyl acrylate 186.67 206.99 204.87
245 methyl methacrylate 183.58 205.73 203.76
246 vinyl propionate 179.18 203.47 203.96
247* 2-hydroxyethyl acrylate 226.93 225.72 221.62
248 methyl acetoacetate 200.66 219.71 217.04

34 QSAR Comb. Sci. 22 (2003)

Quantitative structure property relationship models for the prediction of liquid heat capacity

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

249 valeronitrile 189.82 178.99 176.24
250 cyclopentane 119.78 122.09 137.60
251 2-methyl-1-butene 156.02 160.30 162.67
252 2-methyl-2-butene 152.82 150.52 152.06
253 3-methyl-1-butene 150.42 155.70 158.15
254* 1-pentene 153.18 157.50 161.83
255 cis-2-pentene 151.67 155.50 158.61
256 trans-2-pentene 153.37 155.50 158.61
257 2 2,3-dibromo-2-methylbutane 168.67 188.57 192.84
258 1,5-dichloropentane 214.96 193.90 191.57
259 methyl isopropyl ketone 179.46 179.09 182.43
260 diethyl ketone 191.76 189.24 192.56
261* valeraldehyde 185.76 188.11 191.44
262 n-butyl formate 193.52 212.64 212.63
263 ethyl propionate 199.10 218.96 218.53
264 isobutyl formate 214.35 208.03 208.37
265 isopropyl acetate 199.07 205.98 208.42
266 n-propyl acetate 196.55 213.51 214.51
267 methyl n-butyrate 205.45 218.67 218.53
268* 2-methylbutyric acid 217.00 211.17 211.93
269 isovaleric acid 238.34 204.19 207.34
270 valeric acid 207.35 210.25 213.27
271 diethyl carbonate 212.06 238.37 236.39
272 ethyl lactate 254.59 230.16 227.82
273 1-bromopentane 150.69 183.37 183.18
274 1-chloropentane 187.32 184.33 184.44
275* 1-chloro-3-methylbutane 138.35 180.40 181.55
276 2-chloro-2-methylbutane 133.24 174.31 178.70
277 isopentane 158.40 164.53 165.12
278 neopentane 167.00 151.70 159.31
279 n-pentane 163.64 166.10 168.13
280 2-methyl-1-butanol 226.68 209.23 209.33
281 2-methyl-2-butanol 251.62 190.05 196.12
282* 3-methyl-1-butanol 164.90 204.94 207.14
283 3-methyl-2-butanol 202.52 199.68 201.59
284 1-pentanol 207.78 208.33 210.85
285 2-pentanol 243.98 201.09 204.63
286 3-pentanol 252.51 204.94 207.00
287 methyl sec-butyl ether 192.16 203.08 201.27
288 methyl tert-butyl ether 183.66 188.45 192.54
289* methyl isobutyl ether 190.88 202.26 201.20
290 ethyl propyl ether 199.04 206.14 204.76
291 1,5-pentanediol 225.24 228.21 229.03
292 2-(2-methoxyethoxy)ethanol 280.22 259.00 253.92
293 n-pentyl mercaptan 201.63 195.55 195.86
294 butyl methyl sulfide 199.23 204.24 208.88
295 ethyl propyl sulfide 197.30 191.81 196.09
296* 2-methyl-2-butanethiol 178.95 178.49 182.61
297 n-pentylamine 209.44 213.08 218.22
298 hexafluorobenzene 220.17 227.68 237.94
299 1,2,4-trichlorobenzene 202.48 181.10 185.24
300 m-dibromobenzene 181.38 165.42 167.44
301 m-dichlorobenzene 171.97 163.10 168.05
302 o-dichlorobenzene 185.68 164.75 168.35
303* m-difluorobenzene 135.39 171.68 171.72
304 o-difluorobenzene 135.70 167.18 169.33
305 p-difluorobenzene 134.86 162.46 166.69
306 bromobenzene 159.05 147.94 150.41
307 monochlorobenzene 152.60 143.51 148.35
308 o-chlorophenol 198.53 182.69 180.00
309 fluorobenzene 148.21 147.74 151.21
310* iodobenzene 160.14 161.53 157.62

QSAR Comb. Sci. 22 (2003) 35

 X. Yao et al.

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

311 nitrobenzene 189.62 212.62 196.78
312 benzene 137.87 120.42 129.97
313 m-chloroaniline 210.51 198.08 189.70
314 phenyl mercaptan 174.80 158.13 158.58
315 aniline 210.44 176.67 173.49
316 2-methylpyridine 159.66 159.47 161.72
317* 3-methylpyridine 161.28 157.52 161.41
318 4-methylpyridine 173.01 158.62 161.72
319 adiponitrile 159.47 205.78 201.26
320 methylglutaronitrile 221.58 215.55 208.95
321 dimethyl maleate 260.78 261.45 255.23
322 1-methylcyclopentene 147.14 148.29 156.54
323 3-methylcyclopentene 141.33 150.42 158.72
324* 4-methylcyclopentene 146.77 150.87 160.23
325 cyclohexene 142.31 145.61 155.11
326 2,3-dimethyl-1,3-butadiene 180.09 173.07 174.99
327 1,5-hexadiene 170.61 174.19 178.31
328 1-hexyne 187.01 176.76 178.73
329 2-hexyne 170.76 181.10 179.13
330 3-hexyne 168.74 180.58 179.88
331* cyclohexanone 190.32 184.81 186.93
332 ethyl methacrylate 199.81 233.73 230.40
333 n-propyl acrylate 226.37 236.14 233.25
334 propionic anhydride 233.72 241.98 239.99
335 ethylene glycol diacetate 263.61 262.77 257.19
336 ethylidene diacetate 249.15 252.34 248.24
337 hexanenitrile 223.97 206.34 203.26
338* cyclohexane 162.07 154.38 164.13
339 2,3-dimethyl-1-butene 175.97 183.72 185.03
340 2,3-dimethyl-2-butene 172.96 176.97 176.54
341 3,3-dimethyl-1-butene 179.09 176.92 181.97
342 2-ethyl-1-butene 168.49 194.31 194.68
343 trans-3-methyl-2-pentene 175.03 185.82 184.85
344 1-hexene 178.16 186.75 189.25
345* cis-2-hexene 168.34 197.69 168.83
346 trans-2-hexene 176.97 184.18 186.01
347 cis-3-hexene 177.06 197.69 168.83
348 trans-3-hexene 184.65 183.95 186.01
349 methylcyclopentane 150.72 162.98 170.38
350 2-methyl-1-pentene 175.77 186.74 188.41
351 2-methyl-2-pentene 165.07 181.95 183.55
352* 3-methyl-1-pentene 193.72 186.62 186.03
353 4-methyl-1-pentene 162.96 182.91 185.65
354 4-methyl-cis-2-pentene 173.22 180.08 182.69
355 4-methyl-trans-2-pentene 181.42 180.23 182.69
356 butyl vinyl ether 214.09 220.87 219.94
357 cyclohexanol 218.25 205.53 207.36
358 1-hexanal 211.15 217.54 218.50
359* ethyl isopropyl ketone 207.32 210.38 212.58
360 2-hexanone 213.77 208.63 211.87
361 3-hexanone 217.14 212.55 214.56
362 methyl isobutyl ketone 212.42 204.45 209.70
363 n-pentyl formate 242.31 241.99 241.65
364 n-butyl acetate 228.25 243.33 243.90
365 sec-butyl acetate 230.57 239.07 238.95
366* tert-butyl acetate 236.76 223.06 222.81
367 ethyl n-butyrate 230.41 248.01 247.16
368 ethyl isobutyrate 209.83 244.97 242.97
369 isobutyl acetate 232.55 237.60 237.92
370 n-propyl propionate 225.91 248.44 247.35
371 2-ethyl butyric acid 254.32 244.02 242.69
372 n-hexanoic acid 268.25 240.70 242.48

36 QSAR Comb. Sci. 22 (2003)

Quantitative structure property relationship models for the prediction of liquid heat capacity

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

373* 2-ethoxyethyl acetate 244.75 266.99 264.72
374 cyclohexylamine 216.95 209.96 212.03
375 2,2-dimethylbutane 182.23 185.49 188.06
376 2,3-dimethylbutane 181.25 189.87 189.20
377 n-hexane 192.63 194.80 195.77
378 2-methylpentane 184.90 191.41 192.08
379 3-methylpentane 181.96 195.54 193.46
380* 2-ethyl-1-butanol 259.44 240.88 239.41
381 1-hexanol 219.55 237.77 239.50
382 2-hexanol 244.31 229.64 232.41
383 2-methyl-1-pentanol 259.91 237.14 237.03
384 4-methyl-2-pentanol 248.90 225.31 227.86
385 n-butyl ethyl ether 225.79 234.43 231.96
386 diisopropyl ether 214.32 281.96 281.11
387* di-n-propyl ether 220.21 293.43 291.99
388 methyl tert-pentyl ether 217.28 222.18 221.92
389 diethylene glycol dimethyl 272.85 284.56 279.75
390 dipropylene glycol 346.58 270.88 269.59
391 2-(2-ethoxyethoxy)ethanol 307.75 287.92 284.54
392 n-hexyl mercaptan 230.85 224.36 223.46
393 butyl ethyl sulfide 231.75 221.72 224.20
394* diisopropyl sulfide 202.65 201.97 205.13
395 methyl pentyl sulfide 212.07 234.31 237.26
396 dipropyl sulfide 223.70 221.72 224.20
397 dipropyl disulfide 249.50 239.44 239.20
398 diisopropylamine 228.07 227.24 228.98
399 di-n-propylamine 228.92 237.82 238.52
400 n-hexylamine 270.51 242.98 247.03
401* triethylamine 211.97 239.82 236.49
402 triethanolamine 375.06 308.18 306.28
403 2,4-dichlorobenzotrifluoride 244.74 232.15 226.32
404 p-chlorobenzotrifluoride 218.90 214.85 211.30
405 m-chlorobenzoyl chloride 206.23 219.63 210.85
406 benzoyl chloride 181.15 202.37 192.32
407 benzotrichloride 213.47 200.47 202.15
408* benzotrifluoride 190.80 198.32 196.53
409 benzonitrile 180.55 169.95 169.98
410 benzyl dichloride 188.85 191.32 189.83
411 benzaldehyde 177.85 194.72 187.98
412 salicylaldehyde 233.61 208.43 200.68
413 benzyl chloride 187.41 174.87 176.71
414 o-chlorotoluene 198.51 184.85 185.43
415* p-chlorotoluene 199.43 182.12 182.50
416 p-fluorotoluene 145.48 179.46 181.39
417 m-nitrotoluene 184.96 241.57 226.41
418 o-nitrotoluene 177.89 243.67 230.08
419 o-nitroanisole 257.81 271.03 250.65
420 toluene 157.49 160.93 164.34
421 anisole 203.00 192.40 190.94
422* benzyl alcohol 223.25 211.02 203.63
423 m-cresol 235.69 191.32 189.74
424 benzylamine 220.40 219.82 212.47
425 N-methylaniline 200.35 207.13 204.36
426 m-toluidine 213.39 207.28 203.20
427 o-toluidine 217.20 209.47 206.89
428 1-heptyne 219.27 205.56 205.60
429* n-butyl acrylate 261.14 265.96 263.89
430 isobutyl acrylate 255.76 260.15 257.44
431 n-propyl methacrylate 199.33 262.92 259.45
432 ethylcyclopentane 171.61 198.31 203.07
433 2-ethyl-1-pentene 218.86 221.83 221.99
434 3-ethyl-1-pentene 215.65 218.23 214.82

QSAR Comb. Sci. 22 (2003) 37

 X. Yao et al.

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

435 1-heptene 207.34 216.12 216.40
436* cis-2-heptene 201.57 213.39 213.13
437 trans-2-heptene 208.96 213.39 213.13
438 cis-3-heptene 205.49 213.39 213.13
439 trans-3-heptene 204.63 213.54 213.13
440 methylcyclohexane 171.72 195.12 200.08
441 2-methyl-1-hexene 218.76 215.03 214.91
442 3-methyl-1-hexene 214.72 214.62 213.06
443* 4-methyl-1-hexene 202.07 215.42 215.08
444 2,3,3-trimethyl-1-butene 213.37 206.14 207.64
445 diisopropyl ketone 236.71 239.82 239.07
446 2-heptanone 241.60 238.61 241.00
447 1-heptanal 248.52 247.52 248.45
448 5-methyl-2-hexanone 233.94 233.42 235.34
449 n-butyl propionate 254.71 278.26 278.38
450* ethyl isovalerate 256.17 270.78 269.88
451 isopentyl acetate 254.95 267.33 266.54
452 n-pentyl acetate 263.54 273.15 273.39
453 n-propyl n-butyrate 258.81 277.80 276.97
454 n-heptanoic acid 273.24 271.15 272.93
455 ethyl-3-ethoxypropionate 287.73 301.58 300.63
456 1-bromoheptane 244.42 241.65 241.13
457* N-methylcyclohexylamine 239.76 243.48 245.57
458 2,2-dimethylpentane 206.54 212.41 213.78
459 2,3-dimethylpentane 206.92 221.31 218.52
460 2,4-dimethylpentane 214.24 214.92 216.11
461 3,3-dimethylpentane 206.79 219.25 216.65
462 3-ethylpentane 215.74 227.41 223.48
463 n-heptane 230.42 224.23 223.37
464* 2-methylhexane 216.53 219.74 219.84
465 3-methylhexane 215.88 223.80 221.80
466 2,2,3-trimethylbutane 214.52 211.33 210.95
467 1-heptanol 262.58 267.52 269.24
468 2-heptanol 292.41 259.40 261.23
469 5-methyl-1-hexanol 278.10 262.56 263.25
470 isopropyl-tert-butyl-ether 232.23 236.13 237.20
471* n-heptyl mercaptan 264.09 253.70 252.74
472 butyl propyl sulfide 266.33 251.86 254.21
473 ethyl pentyl sulfide 266.28 251.86 254.21
474 hexyl methyl sulfide 266.28 264.60 267.75
475 1-aminoheptane 264.72 273.28 276.99
476 ethynylbenzene 179.17 178.04 176.89
477 styrene 208.42 189.68 188.73
478* acetophenone 200.32 218.67 210.70
479 p-tolualdehyde 215.20 224.12 215.23
480 methyl benzoate 214.24 256.63 242.68
481 methyl salicylate 252.34 269.15 255.06
482 ethylbenzene 183.60 200.02 200.34
483 m-xylene 182.66 190.75 194.06
484 o-xylene 189.25 192.85 195.93
485* p-xylene 197.75 190.69 194.16
486 phenetole 220.11 223.09 221.09
487 2-phenylethanol 232.08 242.36 235.68
488 2,4-xylenol 253.34 223.51 219.17
489 N,N-dimethylaniline 238.40 231.27 227.38
490 o-ethylaniline 252.44 243.53 236.72
491 2,4,6-trimethylpyridine 233.90 218.39 218.97
492* p-phenetidine 267.40 265.21 256.04
493 diethyl maleate 296.15 320.82 313.44
494 n-butyl methacrylate 288.13 292.81 289.27
495 butyric anhydride 299.98 302.33 301.58
496 diethyl succinate 332.29 332.49 328.83

38 QSAR Comb. Sci. 22 (2003)

Quantitative structure property relationship models for the prediction of liquid heat capacity

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

497 1,1-dimethylcyclohexane 214.81 219.84 222.74
498 ethylcyclohexane 203.43 230.01 232.61
499* 2-ethyl-1-hexene 247.34 251.02 250.91
500 1-methyl-1-ethylcyclopentane 229.79 226.13 227.60
501 1-octene 235.43 245.90 246.15
502 rans-2-octene 244.88 243.21 241.87
503 trans-3-octene 241.36 243.36 242.80
504 trans-4-octene 241.75 243.21 241.87
505 n-propylcyclopentane 211.99 229.59 232.28
506* 2,4,4-trimethyl-1-pentene 223.91 231.76 232.66
507 2,4,4-trimethyl-2-pentene 228.04 225.02 226.94
508 2-ethylhexanal 267.84 278.81 277.46
509 1-octanal 277.82 277.96 278.33
510 2-octanone 274.72 268.84 270.03
511 n-butyl n-butyrate 286.18 308.24 308.82
512 n-hexyl acetate 297.21 303.74 304.70
513* isobutyl isobutyrate 271.00 297.50 295.97
514 n-octanoic acid 332.41 302.06 302.98
515 diethylene glycol ethyl ethe 300.32 351.48 349.98
516 2,2-dimethylhexane 236.50 240.65 240.01
517 2,3-dimethylhexane 255.25 250.03 246.30
518 2,4-dimethylhexane 259.76 247.83 245.09
519 2,5-dimethylhexane 252.56 243.52 242.94
520* 3,3-dimethylhexane 250.25 247.24 243.91
521 3,4-dimethylhexane 253.47 253.60 248.46
522 3-ethylhexane 259.72 256.63 252.56
523 3-ethyl-2-methylpentane 243.85 253.40 248.27
524 3-methyl-3-ethylpentane 246.57 254.00 248.30
525 2-methylheptane 247.85 249.25 248.14
526 3-methylheptane 244.99 253.11 250.53
527* 4-methylheptane 250.91 253.01 250.43
528 4-methylheptane 254.71 254.21 253.26
529 2,2,3-trimethylpentane 241.44 242.94 240.26
530 2,2,4-trimethylpentane 232.22 235.25 235.92
531 2,3,3-trimethylpentane 245.87 245.65 241.60
532 2,3,4-trimethylpentane 233.01 246.94 242.06
533 2,2,3,3-tetramethylbutane 254.33 231.92 231.57
534* di-n-butyl ether 278.23 293.43 291.99
535 di-sec-butyl ether 281.10 284.83 283.70
536 di-tert-butyl ether 270.20 252.92 251.61
537 2-ethyl-1-hexanol 291.04 299.52 300.79
538 1-octanol 325.92 297.87 299.87
539 2-octanol 306.36 289.53 292.11
540 diethylene glycol diethyl ethe 296.67 343.31 342.42
541* diethylene glycol monobutyl et 355.89 347.74 348.72
542 triethylene glycol dimethyl et 372.47 367.25 367.69
543 tetraethylene glycol 441.47 392.35 394.81
544 n-octyl mercaptan 298.05 283.89 282.77
545 tert-octyl mercaptan 287.60 261.09 258.37
546 dibutyl sulfide 282.09 282.51 284.18
547 ethyl hexyl sulfide 299.03 282.67 284.18
548* heptyl methyl sulfide 299.03 295.42 298.25
549 pentyl propyl sulfide 299.03 282.51 284.18
550 dibutyl disulfide 307.25 300.22 300.65
551 di-n-butylamine 298.00 298.15 298.72
552 diisobutylamine 271.50 286.84 286.57
553 n-octylamine 287.48 304.10 308.73
554 cis-propenylbenzene 210.58 217.09 214.41
555* trans-propenylbenzene 211.25 220.48 215.82
556 a-methylstyrene 212.62 219.26 215.23
557 m-methylstyrene 211.62 219.08 215.53
558 o-methylstyrene 211.36 220.58 217.22

QSAR Comb. Sci. 22 (2003) 39

 X. Yao et al.

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

559 p-methylstyrene 212.65 219.40 215.77
560 benzyl acetate 253.30 270.38 258.51
561 ethyl benzoate 244.95 286.99 273.64
562* cumene 217.96 226.47 224.27
563 m-ethyltoluene 213.33 229.38 227.15
564 o-ethyltoluene 218.70 231.39 228.84
565 p-ethyltoluene 216.40 229.67 227.38
566 mesitylene 208.31 219.13 219.85
567 n-propylbenzene 215.43 231.15 229.18
568 1,2,3-trimethylbenzene 216.26 223.47 223.86
569* 1,2,4-trimethylbenzene 213.52 221.29 222.05
570 benzyl ethyl ether 248.44 269.25 260.46
571 1-nonyne 280.06 265.03 263.63
572 butylcyclopentane 254.13 260.81 262.79
573 isopropylcyclohexane 243.58 256.91 257.16
574 1-nonene 281.84 276.41 275.82
575 n-propylcyclohexane 243.96 261.16 262.79
576* diisobutyl ketone 305.70 291.65 291.47
577 1-nonanal 312.82 308.77 310.63
578 n-butyl valerate 331.31 338.98 342.02
579 n-nonanoic acid 358.69 333.12 337.04
580 n-octyl formate 331.39 333.10 335.39
581 3,3-diethylpentane 289.72 288.77 280.60
582 2,2-dimethyl-3-ethylpentane 257.37 274.41 269.49
583* 3-ethyl-2,3-dimethylpentane 278.37 280.22 273.18
584 2,4-dimethyl-3-ethylpentane 256.86 279.06 273.57
585 2,2-dimethylheptane 283.16 269.61 267.68
586 2,6-dimethylheptane 274.31 272.87 271.10
587 3-ethylheptane 274.66 286.28 282.41
588 4-ethylheptane 275.70 285.98 282.30
589 2,3-dimethylheptane 273.28 279.03 275.22
590* 2,4-dimethylheptane 264.10 276.89 273.52
591 2,5-dimethylheptane 267.85 277.12 273.73
592 3,4-dimethylheptane 273.35 282.85 277.71
593 3,5-dimethylheptane 267.77 281.04 276.46
594 4,4-dimethylheptane 266.77 275.74 272.46
595 3-ethyl-2-methylhexane 270.20 282.30 277.43
596 4-ethyl-2-methylhexane 265.17 280.69 276.15
597* 3-ethyl-3-methylhexane 273.35 282.33 276.91
598 3-ethyl-4-methylhexane 273.11 286.22 279.94
599 2,2,3-trimethylhexane 264.94 271.14 267.00
600 2,2,4-trimethylhexane 256.58 267.66 264.66
601 2,3,3-trimethylhexane 269.83 273.72 268.82
602 2,3,4-trimethylhexane 271.49 279.65 273.94
603 2,3,5-trimethylhexane 262.37 273.29 270.30
604* 2,4,4-trimethylhexane 261.37 270.15 266.55
605 3,3,4-trimethylhexane 273.15 277.67 271.54
606 2-methyloctane 281.87 278.91 277.17
607 3-methyloctane 271.73 282.93 280.04
608 4-methyloctane 271.00 282.67 279.85
609 n-nonane 282.60 284.71 283.00
610 2,2,3,3-tetramethylpentane 282.09 266.64 261.07
611* 2,2,3,4-tetramethylpentane 251.44 263.38 258.43
612 2,2,4,4-tetramethylpentane 265.28 254.24 252.92
613 2,3,3,4-tetramethylpentane 264.21 271.79 265.20
614 2,2,5-trimethylhexane 261.48 263.42 261.61
615 2,6-dimethyl-4-heptanol 330.01 311.65 309.93
616 1-nonanol 380.59 328.84 333.80
617 n-nonyl mercaptan 334.13 314.39 314.85
618* butyl pentyl sulfide 332.77 313.65 316.23
619 ethyl heptyl sulfide 332.77 313.65 316.23
620 hexyl propyl sulfide 332.77 313.65 316.23

40 QSAR Comb. Sci. 22 (2003)

Quantitative structure property relationship models for the prediction of liquid heat capacity

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

621 methyl octyl sulfide 332.77 326.71 332.13
622 n-nonylamine 321.56 335.23 341.79
623 tripropylamine 314.03 327.10 327.85
624 m-divinylbenzene 249.65 241.72 236.23
625* dimethyl phthalate 300.96 351.04 331.77
626 diallyl maleate 367.38 357.88 347.35
627 n-butylbenzene 247.85 261.37 257.09
628 sec-butylbenzene 250.95 261.38 258.47
629 tert-butylbenzene 233.22 247.38 244.56
630 1,2,3,4-tetramethylbenzene 236.92 253.83 251.98
631 m-cymene 246.58 255.04 254.59
632* o-cymene 246.79 257.33 256.62
633 p-cymene 241.22 255.17 254.85
634 m-diethylbenzene 248.67 263.19 261.07
635 o-diethylbenzene 256.97 264.80 262.92
636 p-diethylbenzene 255.72 263.48 261.32
637 2-ethyl-m-xylene 272.15 261.36 260.10
638 2-ethyl-p-xylene 267.15 259.96 258.60
639* 3-ethyl-o-xylene 261.09 261.79 260.33
640 4-ethyl-m-xylene 267.52 259.80 258.60
641 4-ethyl-o-xylene 267.91 260.20 258.84
642 5-ethyl-m-xylene 261.31 257.85 256.37
643 isobutylbenzene 251.74 254.96 252.18
644 1,2,3,5-tetramethylbenzene 280.50 251.14 249.79
645 N,N-diethylaniline 297.52 296.69 293.89
646* 2,6-diethylaniline 335.06 308.47 303.87
647 1-decyne 310.42 295.69 295.25
648 n-butylcyclohexane 317.67 292.24 291.35
649 1-cyclopentylpentane 285.76 292.08 291.35
650 1-decene 306.05 307.22 307.47
651 1-decanal 349.03 339.58 343.15
652 2-ethylhexyl acetate 348.77 365.06 367.06
653* isopentyl isovalerate 363.08 354.92 352.64
654 n-decane 313.73 315.53 314.45
655 2-methylnonane 314.73 309.41 306.65
656 3-methylnonane 310.75 313.28 310.11
657 4-methylnonane 313.00 313.17 309.85
658 5-methylnonane 312.69 312.98 309.85
659 3-ethyloctane 306.31 316.79 312.87
660* 4-ethyloctane 314.18 316.53 312.64
661 2,2-dimethyloctane 317.76 299.50 295.71
662 2,3-dimethyloctane 303.65 308.92 305.82
663 2,4-dimethyloctane 292.75 306.58 303.81
664 2,5-dimethyloctane 296.11 306.90 303.89
665 2,6-dimethyloctane 300.29 307.01 304.16
666 2,7-dimethyloctane 297.91 302.76 301.07
667* 3,3-dimethyloctane 299.61 305.90 301.59
668 3,4-dimethyloctane 302.46 312.73 308.75
669 3,5-dimethyloctane 297.25 310.64 307.11
670 3,6-dimethyloctane 299.08 311.03 307.42
671 4,4-dimethyloctane 294.79 305.62 301.22
672 4,5-dimethyloctane 300.78 312.44 308.60
673 4-propylheptane 294.75 316.04 312.49
674* 4-isopropylheptane 296.56 311.74 307.03
675 3-ethyl-2-methylheptane 299.56 312.18 308.38
676 4-ethyl-2-methylheptane 293.09 310.32 305.49
677 5-ethyl-2-methylheptane 297.63 310.52 307.03
678 3-ethyl-3-methylheptane 302.96 312.22 306.61
679 4-ethyl-3-methylheptane 300.84 316.00 311.23
680 3-ethyl-5-methylheptane 295.68 314.63 309.97
681* 3-ethyl-4-methylheptane 301.89 315.81 311.23
682 4-ethyl-4-methylheptane 299.04 311.74 306.30

QSAR Comb. Sci. 22 (2003) 41

 X. Yao et al.

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

683 2,2,3-trimethylheptane 294.97 300.22 295.03
684 2,2,4-trimethylheptane 283.10 296.78 291.89
685 2,2,5-trimethylheptane 286.28 297.09 292.41
686 2,2,6-trimethylheptane 283.95 292.84 288.54
687 2,3,3-trimethylheptane 298.33 302.94 297.14
688* 2,3,4-trimethylheptane 297.91 308.62 303.08
689 2,3,5-trimethylheptane 298.93 306.93 302.02
690 2,3,6-trimethylheptane 292.90 302.66 298.95
691 2,4,4-trimethylheptane 286.51 299.30 294.06
692 2,4,5-trimethylheptane 293.52 306.57 301.87
693 2,4,6-trimethylheptane 282.25 300.22 296.79
694 2,5,5-trimethylheptane 288.81 299.79 294.65
695* 3,3,4-trimethylheptane 300.53 306.79 300.26
696 3,3,5-trimethylheptane 292.57 303.45 297.55
697 3,4,4-trimethylheptane 299.50 306.46 300.10
698 3,4,5-trimethylheptane 301.26 312.70 305.99
699 3-isopropyl-2-methylhexane 306.71 308.02 302.57
700 3,3-diethylhexane 306.22 318.05 310.76
701 3,4-diethylhexane 303.04 319.19 313.71
702* 3-ethyl-2,2-dimethylhexane 293.00 303.54 297.69
703 4-ethyl-2,2-dimethylhexane 281.48 300.59 294.97
704 3-ethyl-2,3-dimethylhexane 302.88 309.09 302.15
705 4-ethyl-2,3-dimethylhexane 299.19 312.57 305.67
706 3-ethyl-2,4-dimethylhexane 298.15 311.75 305.54
707 4-ethyl-2,4-dimethylhexane 299.50 305.89 299.54
708 3-ethyl-2,5-dimethylhexane 290.43 306.02 301.47
709* 4-ethyl-3,3-dimethylhexane 301.83 310.21 302.83
710 3-ethyl-3,4-dimethylhexane 300.77 313.16 304.77
711 2,2,3,3-tetramethylhexane 298.50 294.74 286.99
712 2,2,3,4-tetramethylhexane 296.52 300.15 292.62
713 2,2,3,5-tetramethylhexane 283.28 293.62 287.73
714 2,2,4,4-tetramethylhexane 290.04 289.63 283.26
715 2,2,4,5-tetramethylhexane 282.60 292.46 286.51
716* 2,2,5,5-tetramethylhexane 270.87 282.10 275.48
717 2,3,3,4-tetramethylhexane 305.67 304.05 295.86
718 2,3,3,5-tetramethylhexane 290.61 296.47 289.90
719 2,3,4,4-tetramethylhexane 301.70 303.13 294.86
720 2,3,4,5-tetramethylhexane 294.62 304.61 296.91
721 3,3,4,4-tetramethylhexane 312.94 301.43 292.57
722 2,4-dimethyl-3-isopropylpentan 293.81 304.30 296.10
723* 3,3-diethyl-2-methylpentane 310.24 315.24 306.29
724 3-ethyl-2,2,3-trimethylpentane 310.03 301.31 292.16
725 3-ethyl-2,2,4-trimethylpentane 291.59 299.57 291.84
726 3-ethyl-2,3,4-trimethylpentane 309.94 306.23 297.36
727 2,2,3,3,4-pentamethylpentane 306.63 292.46 284.08
728 2,2,3,4,4-pentamethylpentane 297.60 288.65 282.02
729 1-decanol 419.11 359.49 365.84
730* di-n-pentyl ether 355.80 354.10 356.17
731 isodecanol 386.48 353.15 357.62
732 tetraethylene glycol dimethyl 400.67 450.89 458.74
733 n-decyl mercaptan 373.33 344.81 346.01
734 butyl hexyl sulfide 366.34 344.55 348.56
735 ethyl octyl sulfide 366.34 344.55 348.56
736 heptyl propyl sulfide 366.34 344.55 348.56
737* methyl nonyl sulfide 366.34 357.60 364.04
738 dipentyl sulfide 366.34 344.55 348.56
739 dipentyl disulfide 358.62 361.83 364.49
740 n-decylamine 348.17 366.13 373.89
741 n-butyl benzoate 292.35 348.65 337.54
742 n-pentylbenzene 290.80 292.54 287.71
743 1-undecyne 340.05 326.18 325.96
744* 2-ethylhexyl acrylate 380.79 388.48 390.90

42 QSAR Comb. Sci. 22 (2003)

Quantitative structure property relationship models for the prediction of liquid heat capacity

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

745 1-undecene 326.54 338.03 339.49
746 1-cyclopentylhexane 317.06 323.72 323.17
747 pentylcyclohexane 318.37 323.88 323.17
748 1-undecanal 362.42 370.70 376.63
749 n-undecane 345.10 346.27 345.15
750 1-undecanol 438.91 390.70 399.75
751* undecyl mercaptan 400.50 375.78 378.99
752 butyl heptyl sulfide 399.06 375.82 381.65
753 decyl methyl sulfide 399.06 389.04 397.99
754 ethyl nonyl sulfide 399.06 375.82 381.65
755 octyl propyl sulfide 399.06 375.82 381.65
756 1,2-dimethylnaphthalene 286.16 275.80 260.63
757 1,3-dimethylnaphthalene 284.16 274.06 257.27
758* 1,4-dimethylnaphthalene 288.40 275.75 260.25
759 1,6-dimethylnaphthalene 279.97 274.20 257.47
760 1,7-dimethylnaphthalene 280.02 274.20 257.47
761 1-ethylnaphthalene 287.58 285.34 270.48
762 2-ethylnaphthalene 272.80 283.83 268.25
763 diethyl phthalate 374.95 406.99 392.15
764 m-diisopropylbenzene 299.26 314.03 308.62
765* p-diisopropylbenzene 314.53 314.16 308.95
766 n-hexylbenzene 318.15 323.56 320.03
767 1,2,3-triethylbenzene 288.62 328.55 327.71
768 1,2,4-triethylbenzene 288.62 327.40 325.91
769 1,3,5-triethylbenzene 286.42 325.30 324.00
770 dibutyl maleate 432.52 444.20 442.54
771 1-dodecyne 369.23 357.14 358.57
772* 1-dodecene 358.08 369.08 372.05
773 1-cyclopentylheptane 348.23 355.15 355.22
774 1-cyclohexylhexane 351.58 355.63 355.22
775 1-dodecanal 452.28 401.90 408.64
776 n-dodecane 376.59 377.55 378.56
777 di-n-hexyl ether 418.72 415.95 421.66
778 1-dodecanol 493.25 421.81 432.36
779* diethylene glycol di-n-butyl e 463.35 465.01 475.27
780 n-dodecyl mercaptan 417.03 406.73 410.65
781 butyl octyl sulfide 430.72 407.09 413.46
782 decyl ethyl sulfide 430.72 407.09 413.46
783 dihexyl sulfide 430.70 407.09 413.46
784 methyl undecyl sulfide 430.70 420.31 431.89
785 nonyl propyl sulfide 430.70 407.09 413.46
786* dihexyl disulfide 408.17 424.12 428.81
787 tri-n-butylamine 420.02 418.56 425.63
788 1-propylnaphthalene 296.37 316.59 300.52
789 2-propylnaphthalene 297.89 314.71 298.13
790 n-heptylbenzene 352.82 354.89 352.44
791 1-tridecyne 398.96 388.18 389.75
792 1-tridecene 412.38 400.35 403.40
793* 1-cyclopentyloctane 378.43 386.80 387.22
794 1-cyclohexylheptane 381.14 387.27 387.22
795 1-tridecanal 440.47 433.24 442.11
796 n-butyl nonanoate 463.22 463.23 475.44
797 methyl dodecanoate 455.19 463.25 476.05
798 n-tridecane 407.21 408.66 409.64
799 butyl nonyl sulfide 462.73 438.59 446.40
800* decyl propyl sulfide 462.73 438.59 446.40
801 dodecyl methyl sulfide 462.73 451.81 465.84
802 ethyl undecyl sulfide 462.73 438.59 446.40
803 1-tridecanethiol 453.29 437.91 443.43
804 2-butylnaphthalene 323.91 345.44 328.57
805 n-octylbenzene 387.29 386.07 383.59
806 1,2,3,4-tetraethylbenzene 394.46 392.19 390.39

QSAR Comb. Sci. 22 (2003) 43

 X. Yao et al.

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

807* 1,2,3,5-tetraethylbenzene 393.56 390.66 386.65
808 1,2,4,5-tetraethylbenzene 392.64 390.66 388.14
809 1-cyclohexyloctane 411.75 418.77 419.18
810 n-tetradecane 438.12 440.00 441.52
811 butyl decyl sulfide 493.42 469.87 479.00
812 dodecyl ethyl sulfide 493.42 469.87 479.00
813 diheptyl sulfide 493.42 469.87 479.00
814* methyl tridecyl sulfide 493.42 483.25 499.49
815 propyl undecyl sulfide 493.42 469.87 479.00
816 1-tetradecanethiol 479.86 469.19 475.85
817 diheptyl disulfide 456.31 486.82 491.85
818 1-pentylnaphthalene 353.95 378.27 363.41
819 2-pentylnaphthalene 360.66 376.77 360.72
820 n-nonylbenzene 426.78 417.14 415.82
821* nonylphenol 473.00 443.57 442.10
822 1-pentadecyne 454.96 450.56 454.59
823 1-pentadecene 479.35 463.04 467.94
824 1-cyclopentyldecane 437.47 450.10 450.70
825 n-pentadecane 470.02 471.35 473.01
826 butyl undecyl sulfide 523.81 501.44 509.70
827 dodecyl propyl sulfide 523.81 501.44 509.70
828* ethyl tridecyl sulfide 523.81 501.60 511.16
829 methyl tetradecyl sulfide 523.81 514.98 531.17
830 1-pentadecanethiol 507.69 500.61 506.33
831 1-n-hexylnaphthalene 425.40 409.36 393.81
832 dibutyl phthalate 492.41 531.59 518.12
833 n-decylbenzene 457.48 448.62 447.42
834 1-hexadecyne 482.60 481.97 484.82
835* n-decylcyclohexane 464.79 481.90 481.30
836 1-cyclopentylundecane 466.30 481.74 481.30
837 1-hexadecene 484.19 494.61 498.10
838 n-hexadecane 501.90 502.85 502.62
839 di-n-octyl ether 562.61 540.78 549.98
840 butyl dodecyl sulfide 552.68 533.18 542.20
841 ethyl tetradecyl sulfide 553.14 533.18 542.20
842* methyl pentadecyl sulfide 552.68 546.56 563.25
843 dioctyl sulfide 552.68 533.18 542.20
844 propyl tridecyl sulfide 552.68 533.18 542.20
845 1-hexadecanethiol 532.28 532.03 537.11
846 dioctyl disulfide 503.36 549.46 552.96
847 n-undecylbenzene 496.30 480.35 478.49
848 1-heptadecyne 509.82 513.39 515.55
849* 1-cyclopentyldodecane 493.73 513.40 511.11
850 1-heptadecene 545.01 525.97 528.47
851 n-heptadecane 551.90 534.43 532.28
852 butyl tridecyl sulfide 580.88 564.75 572.28
853 ethyl pentadecyl sulfide 580.88 564.75 572.28
854 hexadecyl methyl sulfide 580.88 578.13 594.39
855 propyl tetradecyl sulfide 580.88 564.75 572.28
856* n-dodecylbenzene 531.03 511.68 508.33
857 oleic acid 684.32 604.47 612.05
858 dibutyl sebacate 645.20 645.81 639.37
859 1-cycopentyltridecane 521.38 545.20 539.56
860 1-cyclohexyldodecane 526.93 545.36 539.56
861 1-octadecene 577.72 557.54 557.92
862 dinonyl ether 600.12 603.62 609.05
863* 1-n-nonylnaphthalene 507.01 503.24 483.28
864 n-tridecylbenzene 561.44 543.41 537.20
865 methyl oleate 643.39 640.20 647.39
866 1-cyclohexyltridecane 553.83 577.25 566.86
867 1-nonadecene 602.71 589.12 585.99
868 1-n-decylnaphthalene 538.00 534.64 511.77

44 QSAR Comb. Sci. 22 (2003)

Quantitative structure property relationship models for the prediction of liquid heat capacity

Table 1. (cont.)

Number NAME HEATCAP MLR RBFNNs

869 1-phenyltetradecane 526.12 574.67 562.92
870* 1-cyclopentylpentadecane 573.09 608.57 592.58
871 1-cyclohexyltetradecane 578.96 608.89 592.58
* test set

In which, hj is the notation for the output of the jth RBF unit. 3 Results and Discussion
For the jth RBF cj and rj are the center and width
respectively. The operation of the output layer is linear, A great number of molecular descriptors, which encode the
which is given in equation 2 constitutional, topological and quantum chemical features
of the molecules, were calculated to describe the molecular
X
nh
structures. Forward stepwise regression routine implement-
yk x† ˆ wkj hj x† ‡ bk 2†
jˆ1
ed in SPSS is used to develop the linear model for the
prediction of liquid heat capacity using calculated molecular
descriptors. The best linear model contains six molecular
Where yk is the kth output unit for the input vector x, wkj is descriptors. They are one constitutional, three topological
the weight connection between the kth output unit and the and two quantum chemical descriptors. The one constitu-
jth hidden layer unit and bk is the bias. tional descriptor used is sum of atomic Sanderson electro-
From Eq (1) and Eq (2), one can see that designing a negativities, which is related to the constitution of a
RBFNN involves selecting centers and the corresponding molecule. The topological descriptors used are connectivity
number of hidden layer units, and determining width and index chi-1, connectivity index chi-2, and average connec-
weights. There are various ways for selecting the centers, tivity index chi-0. They describe the size, branching and
such as random subset selection, K-means clustering, composition of a molecule and relate to the dispersion
orthogonal least squares learning algorithm, RBF-PLS interaction among molecules. The two quantum descriptors
[19], etc. The widths of the radial basis function can be are maximum negative charge and mean absolute charge.
chosen either as the same for all the units or different value These two descriptors are correlated with the electrostatic
for each unit. In this paper, considerations were limited to and hydrogen bonding interactions among molecules. The
the Gaussian functions with a constant width for all units. A regression parameters of the best six-descriptor correlation
forward subset selection routine [20, 21] was used to select model are gathered in Table 2. This model produced a
the centers from training set samples. The adjustment of the standard error of 18.36 heat capacity unit and a correlation
connection weights between hidden layer and output layer is coefficient of 0.986. RMS (Root Mean Square Error) was
performed using a least-squares solution after the selection calculated as 18.827 heat capacity unit. Figure 2 shows the
of centers and width of radial basis functions. predicted vs. experimental liquid heat capacities.
The overall performance of RBFNNs is evaluated in After the establishment of a linear model, RBFNN was
terms of root mean squared error (RMS) according to the used to develop a non-linear model based on the same
following equation: subset of descriptors. The RBFNNs has six inputs (a set of six
v molecular descriptors), one output layer unit (liquid heat
uP
2 capacity) and one hidden layer of nh units. Such a RBFNN
u ns
u y ^ yk can be designed as 6-nh-1 net, indicating the number of units
t iˆ1 k
RMS ˆ 3† in input layer, hidden layer and output layer respectively. A
ns
RBFNN is completely specified by choosing the following
parameters:
Where yk is the desired output and yÃk is the actual output of
* The number nh of radial basis functions
the network, ns is the number of compounds in analyzed set.
* The center cj and width rj of each radial basis function
* The connection weight wkj between jth hidden layer unit
2.4 Radial Basis Function Neural Networks and kth output unit
implementation and computation environment
The number of radial basis functions (the hidden layer units)
All calculation programs implementing RBFNNs were nh greatly influences the performance of a RBFNN. If the
written in M-file based on basis MATLAB script for radial number is too low, the network may not calculate a proper
basis function neural networks [20, 21]. All scripts were estimation of the data. On the other hand, if too many
compiled using MATCOM compiler running Redhat Linux hidden layer units are used, the network tends to overfit the
6.0 operating system on a Pentium 266 PC with 128M training data. In this paper, the radial basis functions were
RAM. added one by one and terminated if no performance of the

QSAR Comb. Sci. 22 (2003) 45

 X. Yao et al.

Table 2. Descriptors, coefficients, standard error, and T-values for the linear model
Chemical Meaning Descriptor Coefficient Std. Error t value
Intercept ( Constant) 180.939 17.486 10.348
sum of atomic Sanderson electronegativities SE 6.509 .243 26.755
connectivity index chi-1 X1 33.924 1.876 18.079
maximum negative charge QMIN 191.878 14.433 13.294
average connectivity index chi-0 X0A 230.778 18.937 12.187
connectivity index chi-2 X2 11.013 1.085 10.146
mean absolute charge QMEAN 157.903 18.165 8.693
R (correlation coefficient) 0.986
F value 4738.46
Std 18.67

Figure 2. The predicted vs. experimental liquid heat capacity using MLR

networks was improved by adding a new basis function. The increasing the model complexity is continued till some
centers of RBFNNs are determined with the forward subset criterion such as GCV stops increasing. The criterion of the
selection method proposed by Orr [20 ± 21] . The advantages selection used here is an approximation of the leaving-one-
of this method over other center selection ones is that it can out (LOO) cross-validation methods, according to the
determine the number of hidden layer units simultaneously equation below:
and there is no need to fix the number of hidden layer units
in advance. This method has also a tractable model order yP diag P†† P ^
^ 2
y
s 2LOO ˆ 4†
selection and goes through a process of selecting a subset of p
radial basis functions from a large set of candidates (training
set samples). The model starts empty; the radial basis Where yà is the output of the network, P is the projection
function is added one by one by selecting the candidate from matrix, which can be computed by P ˆ Ip ZZt from the
sample list, followed by testing if the added center reduces outputs matrix Z of hidden layer units and the unit matrix Ip
the sum of squared errors compared with other samples, with dimension p, p is the pattern number in training sets.
until the sum of squared errors reached its minimum and not The LOO cross-validation method was used to prevent the
further changed. This process of adding hidden units and network from overfitting.

46 QSAR Comb. Sci. 22 (2003)

Quantitative structure property relationship models for the prediction of liquid heat capacity

Figure 3. The optimization of the width of RBFNNs Figure 4. The predicted vs. experimental liquid heat capacity
using RBFNNs

After the selection of the centers and number of hidden effect on the performance of RBFNNs if width exceeds 3. So
layer units, the connection weights can be easily calculated we select the optimal width from 1.5 to 3 using an increment
by linear least square methods. of 0.1. Each minimum error on LOO cross-validation was
plotted versus the width (Figure 3) and the minimum was
w ˆ yZ'(ZZ') 1
(5) chosen as the optimal value. Finally, in our study the two
parameters r and nh are equal to 1.9 and 19 respectively.
Where y is the matrix of training example targets, Z is the The selected centers and their distribution among training
matrix of hidden layer unit outputs, Z' is the transpose of samples are listed in Table 3. As can be seen the selected
matrix Z and w is the weight matrix connection hidden layer centers correspond very well with the distribution of train-
and output layer. ing set samples. With the best network, a test was carried out
The optimal width was determined by experiments with a using the test set, and the results obtained are given in
number of trials by taking into account of the model Table 1. The plot of predicted values versus experimental
selection criterion: a width smaller than 1.5 gives poor data is shown in Figure 4. The network gives RMS of 17.141
prediction ability, varying the width indicates width has little for the whole set, 16.857 for the training set, and 18.744 for
the prediction set. These results are generally better than
those obtained with MLR method. Analysis of the obtained
Table 3. A full list of centers selected for RBFNNs results indicates that the RBFNN model we proposed can
well represent the structural-property relationships of these
No Name
compounds, and that we can use only the parameters
858 dibutyl sebacate calculated from structures as molecular descriptors for
832 dibutyl phthalate predicting the liquid heat capacities of studied compounds.
865 methyl oleate Compared with previous work reported for alkanes [9, 10],
859 1-cycopentyltridecane
864 n-tridecylbenzene
the RMS values seem to be more important in this study, this
146 decafluorobutane can be explained by diversity of the data set used in our
20 methylchloride investigation, they are not only limited to alkanes. The
82 ethylamine model obtained from this study covers almost all types of
159 methacrylic acid molecules. It is a generally applicable model in the
110 acetone prediction of heat capacity. The result is therefore very
831 1-n-hexylnaphthalene
satisfactory.
403 2,4-dichlorobenzotrifluoride
69 acetic acid
757 1,3-dimethylnaphthalene
857 oleic acid 4 Conclusion
97 methyl chloroacetate
85 hexafluoropropylene QSPR models for the prediction of liquid heat capacity for a
469 5-methyl-1-hexanol
diverse set of organic compounds using RBFNNs based on
301 m-dichlorobenzene
descriptors calculated from molecular structure alone have

QSAR Comb. Sci. 22 (2003) 47

 X. Yao et al.

been developed. The models proposed could identify and [8] Yaws, C. L. Chemical properties handbook, McGraw-Hill,
provide some insight into what structural features are New York, 1999.
related to liquid heat capacities of these compounds. [9] Gakh, A. A., Gakh, E. G., Sumpter, B. G., Noid, D. W. Neural
network-Graph Theory approach to the prediction of the
RBFNNs were proved to be a useful tool in the prediction physical properties of organic componds. J. Chem. Inf.
of heat capacity. The training procedure is also simple when Comput. Sci. 34, 832 ± 839.(1994).
using RBFNNs because there are fewer parameters having [10] Liu, S., Cai, S., Cao, C., Li, Z Molecular electronegative
to be optimized: the width of radial basis function and the distance vector (MEDV) related to 15 properties of alkanes,
number of units in the hidden layer. Furthermore the J. Chem. Inf. Comput. Sci, 40, 1337 ± 1348, (2000).
proposed approach can also be extended in other QSPR/ [11] Yao, X. J., Zhang, X. Y., Zhang, R. S., Liu, M. C., Hu, Z. D.,
Fan, B. T. Prediction of enthalpy of alkanes by the use of
QSAR investigations.
radial basis function neural network. Comput. Chem. 25,
475 ± 483, (2001).
[12] Yao, X. J., Zhang, X. Y., Zhang, R. S., Liu, M. C., Hu, Z. D.,
Acknowledgement Fan, B. T. Radial basis function neural networks based QSPR
for the prediction of critical pressure of substituted benzenes.
The authors thank the Association Franco-Chinoise pour la Comput. Chem. 26, 159 ± 169, (2002).
[13] Yao, X. J., Zhang, X. Y., Zhang, R. S. ; Liu, M. C. ; Hu, Z. D.,
Recherche Scientifique & Technique (AFCRST) for sup-
Fan, B. T. Prediction of gas chromatography indices by the
porting this study (Programme PRA SI 00-05). Special use of radial basis function neural network. Talanta, 57, 297 ±
thanks were given to Prof.Todeschini and other members in 306, (2002).
Milano Chemometrics and QSAR Research Group for [14] Yao, X. J., Yawei Wang, Zhang, X. Y., Zhang, R. S., Liu,
providing Dragon package for use in this research. The M. C., Hu, Z. D., Fan, B. T. Radial basis function neural
authors would also like to thank Mark J. Orr for providing networks based QSPR for the prediction of critical temper-
his MATLAB routines to the scientific community. atures Chemometr. Intell. Lab. Syst, 62, 217 ± 225, (2002).
[15] Yao, X. J., Liu, M. C., Zhang, X. Y., Hu, Z. D., Fan, B. T.
Radial basis function neural networks based QSPR for the
prediction of Henrys law constant, Anal.Chim.Acta, 462,
References 101 ± 117, (2002).
[16] Xiang, Y. H., Liu, M. C., Zhang, X. Y., Zhang, R. S., Hu,
[1] Katritzky, A. R., Maran, U., Lobanov, V. S., Karelson, M. Z. D., Fan, B. T., Doucet, J. P., Panaye, A. Quantitative
Structurally Diverse Quantitative Structure-Property Rela- Prediction of Liquid Chromatography Retention of N-Ben-
tionship Correlations of Technologically Relevant Physical zylideneanilines Based on Quantum Chemical Parameters
Properties. J. Chem. Inf. Comput. Sci. 4, 1 ± 18 (2000). and Radial Basis Function Neural Network J. Chem. Inf.
[2] Katritzky, A. R., Petrukhin, R., Tatham, D., Basak, S., Comput. Sci, 42, 592 ± 597, (2002).
Benfenati, E., Karelson, M., Maran, U. Interpretation of [17] HyperChem 4.0, Hypercube, Inc, 1994.
Quantitative Structure-Property and Activity Relationships. [18] Todeschini, R. Dragon software for the calculation of the
J. Chem. Inf. Comput. Sci. 41, 679 ± 685 (2001). molecular descriptors, Rel. 1.1 for Windows. Milano, 2000.
[3] Karelson, M. Molecular Descriptors in QSAR/QSPR; John [19] Walczak, B., Massart, D. L. Local modeling with radial basis
Wiley & Sons: New York, 2000. function networks. Chemometr. Intell. Lab. Syst. 50, 179 ± 198,
[4] Todeschini, R., Consonni, V. Handbook of Molecular De- (2000).
scriptors; Wiley-VCH: Weinheim, Germany, 2000. [20] Orr, M. J. L. Introduction to Radial basis function networks,
[5] Devillers, J., Balaban, A. T., Eds. Topological Indices and centre for cognitive science, Edinburgh University, 1996.
Related Descriptors in QSAR and QSPR; Gordon and [21] Orr, M. J. L. MATLAB routines for subset selection and
Breach: Amsterdam, The Netherlands, 1999. ridge regression in linear neural networks, Centre for
[6] Estrada, E., Uriarte, E. Recent advances on the role of cognitive science, Edinburgh University, 1996.
topological indices in drug discovery research. Curr. Med.
Chem. 8, 1699 ± 1714 (2001). Received on: June 14, 2002
[7] Karelson, M., Lobanov, V. S., Katritzky, A. R. Quantum-
chemical descriptors in QSAR/QSPR studies. Chem. Rev. 96,
1027 ± 1043 (1996).

48 QSAR Comb. Sci. 22 (2003)