Phytochemistry 93 (2013) 182–191

Contents lists available at SciVerse ScienceDirect

Phytochemistry
journal homepage: www.elsevier.com/locate/phytochem

Classification of flavonoid compounds by using entropy of information theory

Gloria Castellano a,⇑, Juan Luis González-Santander a, Ana Lara a, Francisco Torrens b
a
Facultad de Ciencias Experimentales, Universidad Católica de, Valéncia San Vicente Mártir, Guillem de Castro-94, E-46001 Valencia, Spain
b
Institut Universitari de Ciéncia Molecular, Universitat de València, Edificid’Instituts de Paterna, P.O. Box 22085, E-46071 Valencia, Spain

a r t i c l e i n f o a b s t r a c t

Article history: A total of 74 flavonoid compounds are classified into a periodic table by using an algorithm based on the
Received 13 June 2012 entropy of information theory. Seven features in hierarchical order are used to classify structurally the
Received in revised form 8 March 2013 flavonoids. From these features, the first three mark the group or column, while the last four are used
Available online 3 May 2013
to indicate the row or period in a table of periodic classification. Those flavonoids in the same group
and period are suggested to show maximum similarity in properties. Furthermore, those with only the
Keywords: same group will present moderate similarity. In this report, the flavonoid compounds in the table, whose
Information entropy
experimental data in bioactivity and antioxidant properties have been previously published, are related.
Molecular classification
Antioxidant activity
Ó 2013 Elsevier Ltd. All rights reserved.
Flavonoid
Polyphenol

1. introduction a chain-propagating radical (Croft, 1998). Another pathway of

The aim of the present report is to develop the learning poten-
tialities of our program of molecular structural classification,
which is applied to flavonoids. The molecules with similar chemi-
cal structures are described. We have extended the study of gen-
eral approaches to the processing of structural information. In
this way, the antioxidant activity of any flavonoid can be predicted,
depending on its molecular structure as compared with those that
have been already classified.
Flavonoids are polyphenolic compounds that are secondary
metabolites exclusively from vegetal origin. All flavonoids arise
from the combined biosynthesis of the shikimic acid and ace-
tate–malonate pathways. There is much interest in the biological
effects of flavonoids because they exhibit antioxidant, anti-inflam-
matory, anti-tumor, anti-viral and anti-allergical properties (Croft,
1998; Dumay et al., 2004; Khadem and Marles, 2010; Lacikova
et al., 2009; Pietta, 2000; Yang et al., 2001). Moreover, flavonoids
may play an important role in atherosclerosis prevention and ther-
apy (Yi et al., 2001).
Flavonoids (cf. Fig. 1) are C15 compounds composed of two phe-
nolic rings connected by a three-carbon unit.
These compounds can act as antioxidants by a number of poten-
tial pathways. The most important one is likely to be by free-rad-
ical scavenging, in which the polyphenol can break the chain
reaction of free radicals (Pietta, 2000). The resulting radical formed
on the polyphenol has to be stable so as to prevent it from acting as
⇑ Corresponding author. Tel.: +34 963544431; fax: +34 963543274.
E-mail address: gloria.castellano@ucv.es (G. Castellano).
apparent antioxidant action of the flavonoids is chelation of cata-
lytic metal ions, as copper or iron, that may prevent their involve-
ment in Fenton-type and Haber–Weiss reactions, which can
generate hydroxyl radicals. The ability of flavonoids to react with
metal ions may also render them pro-oxidant: reduce Cu2+ to Cu+
and hence allow the formation of initiating radicals (Cao et al.,
1997; Halliwell et al., 1995).
The radical-scavenging antioxidants inhibit the free-radical-
mediated oxidation of lipids, proteins and DNA, which is involved
in diseases. Phenolic compounds act as antioxidants by inhibiting
enzymes involved in radical generation (Anouar et al., 2009; De-
wick, 2002; Fukumoto and Mazza, 2000).
Among the several methods designed to measure the activity of
compounds, there are such as the lipid peroxidation inhibition
capacity (LPIC) assay (Zhang et al., 2007), 1,1-diphenyl-2-pic-
rylhydrazyl (DPPH) scavenging (Brand-Williams et al., 1995), cyclic
voltammetry (Simic et al., 2007) and using the induction period
method (Kadoma and Fujisawa, 2008). The antioxidants act syner-
gistically together with other antioxidants in vivo (Omata et al.,
2008).
The main structural features of flavonoids have to pay attention
to some structural requirements for efficient antioxidant potency:
an ortho-dihydroxy (catechol) structure on the B ring, which con-
fers greater stability to aroxyl radicals, possibly through hydrogen
bonding, and which participates in electron delocalization; 2,3-
double bond in conjugation with a 4-keto function provides elec-
tron delocalization from the B ring; hydroxyl groups at positions
3 and 5 provide hydrogen bonding to the keto group (Bors et al.,
1990; Croft, 1998; Patra and Chua, 2011). Even, the occurrence of

0031-9422/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.phytochem.2013.03.024
 G. Castellano et al. / Phytochemistry 93 (2013) 182–191 183

tion of molecule i or j. The hierarchical order of the seven structural

features is expressed by their corresponding weights. For instance,
for all ak = 0.5 these weights are 0.5, 0.25, 0.125, 0.0625, 0.03125,
0.015625 and 0.0078125.
The grouping algorithm uses the stabilized similarity matrix ob-
tained by applying the max–min composition rule defined:
 SÞ ¼ max½minðr ; s Þ
ðRo ð2Þ
ij ik kj
k k

where R  ¼ ½r  and S ¼ ½s  are two similarity matrices and ðRo

 SÞ is
ij ij ij
the (i,j)th element of the matrix Ro  S (Kaufmann, 1975). It can be
shown that when applying this rule iteratively so that
Rðn 
 þ 1Þ ¼ RðnÞo 
R, there is an integer n such that:

RðnÞ  þ 1Þ ¼ . . . The resulting matrix RðnÞ
¼ Rðn  is called the stabi-
Fig. 1. Basic structure of flavonoids.
lized similarity matrix, which importance lies in the fact that in
substituents with positive mesomeric and inductive electronic ef- the classification process, it will generate a partition into disjoint

classes. It is used and designated by RðnÞ ¼ ½rij ðnÞ. The grouping rule
fects is also needed. The presence of certain hydroxyl groups on
the flavonoid rings enhances antioxidant activity; glycosylation follows: i and j are assigned to the same class if rij(n) P b, where b is
of flavonoids diminishes their activity when compared to the cor- the grouping level of the classification. Class of i noted ^i is the set of
responding aglycones; alkyl and mesomeric methoxy groups pos- species j that satisfies the rule rij(n) P b. The matrix of classes is:
sessing positive inductive effects show such function, and O- 
RðnÞ ¼ ½^r^i^j  ¼ maxðr st Þðs 2 ^i; t 2 ^jÞ ð3Þ
methylation may reflect steric effects that perturb planarity and s;t
decrease antioxidant activity and hydrophobicity. Flavonoids with
where s stands for any index of a species belonging to the class ^i
gallate or cinnamate moiety reflect the contribution from gallic or
(similarly for t and ^j). Rule (3) means finding the largest similarity
cinnamic acid, and increase antioxidant action (Rice-Evans et al.,
index between species of two different classes.
1996). Conjugation between the A- and B rings permits a resonance
The information entropy associated with the matrix of similar-
effect of the aromatic nucleus that lends stability to the flavonoid  is:
ity R
radical.
 ¼ X X
In an earlier publication, it was examined the molecular classi- hðRÞ r ij ln r ij  ð1  rij Þ lnð1  r ij Þ ð4Þ
fication of 33 phenolic compounds derived from cinnamic and ben- i;j i;j
zoic acids (Castellano et al., 2012). In the present report, 74
flavonoids: anthocyanidins, flavanes, flavones, flavonols and flava- In the classification algorithm, each hierarchical tree corre-
nones have been clustered. sponds to a dependence of entropy on the grouping level and an
h–b diagram can be obtained (White, 1989). The equipartition con-
jecture of entropy production is proposed as a selection criterion,
2. Calculation
among different variants, resulting from classification among hier-
archical trees. According to this conjecture, the best configuration
The computational method was applied to polyphenols (Castell-
of a dendrogram is that in which entropy production is most uni-
ano et al., 2012). First step in quantifying the concept of similarity
formly distributed.
for molecules of flavonoid compounds is to list the most important
Learning procedures similar to stochastic methods are imple-
moieties with respect to the antioxidant activity. Properties vector
i = hi , i , . . . i , . . . i should be associated with each flavonoid i, whose mented. Consider a given partition into classes as good from prac-
1 2 k
tical observations. This corresponds to a reference similarity
components correspond to a number of characteristic groups in the
matrix S ¼ ½sij  obtained for an arbitrary number of fictitious prop-
molecule, in a hierarchical order according to the expected impor-
erties. Next, consider the same set of species as in the good classi-
tance of their antioxidant potency. Components ik are ‘‘1’’ or ‘‘0’’,
fication and the actual properties. The similarity of degree rij is
according to experimental conclusions of antioxidant power to  The number of
then computed with Eq. (1) giving the matrix R.
structural variations in the flavonoid: index i1 = 1 denotes either  
properties for R and S is:
a cinnamic or benzoic ester group at 3-position on ring C (Fig. 1),
i2 = 1 signifies an O-dihydroxy (catechol) structure on the B ring X 1  r ij X r ij
D¼ ð1  r ij Þ ln  ri;j ln 80 6 r ij ; sij 6 1 ð5Þ
for electron delocalization, i3 = 1 indicates the presence of two hy- 1  s ij sij
ij ij
droxyl groups at positions 3 and 5, i4 = 1 means the presence of a
2,3-double bond in conjugation with a 4-keto function, providing The definition was suggested by that introduced in information
electron delocalization from the B ring , i5 = 1 represents both a theory by Kullback (1959) to measure the distance between two
2,3-double bond and 4,5-double bond that cause electron delocal- probability distributions. Algorithm result is a set of weights allow-
ization from the B ring, i6 = 1 shows the absence of alkoxyl and ing adequate classification. Such a procedure was applied in the
glycoxyl groups on ring B, and i7 = 1 indicates the absence of alk- synthesis of complex flow sheets using information entropy (Iord-
oxyl and glycoxyl groups on ring A (Table 1). ache et al., 1993).
Let us denote by rij (0 6 rij 6 1) the similarity index of two phe-
nolic compounds associated with the i and j vectors, respectively. 2.1. Calculation outcomes
The relation of similitude is characterized by a similarity matrix
 ¼ ½r . The similarity index between two phenolic compounds
R The matrix of Pearson correlation coefficients has been calcu-
ij
i = hi , i , . . . i , . . . i and j=hj , j , . . . j , . . .i is defined as: lated between each pair of vector properties hi1, i2, i3, i4, i5, i6, i7i
1 2 k 1 2 k
X of the 74 flavonoids. The Pearson intercorrelations are computed
r ij ¼ t k ðak Þk ðk ¼ 1; 2; . . .Þ ð1Þ for the partial correlation diagram, which contains high partial cor-
k
relations (r P 0.75), medium partial correlations (0.50 6 r < 0.75),
where 0 6 ak 6 1 and tk = 1 if ik = jk, but tk = 0 if ik – jk. This defini- low partial correlations (0.25 6 r < 0.50) and zero partial correla-
tion assigns a weight (ak)k to any property involved in the descrip- tions (r < 0.25). Pairs of flavonoid compounds with high partial
184 G. Castellano et al. / Phytochemistry 93 (2013) 182–191

Table 1
Chemical structures of flavonoids in Fig. 1 and vector properties.

No. Flavonoid 2 3 4 5 6 7 8 20 30 40 50 60 Vector properties

1 Acacetin – H @O OH H OH H H H OMe H H h0001001i
2 Apigenin – H @O OH H OH H H H OH H H h0001011i
3 Artemetin – OMe @O OH OMe OMe H H OMe OMe H H h0001000i
4 Astragalin – OGlu @O OH H OH H H H OH H H h0001011i
5 Axillarin – OMe @O OH OMe OH H H OH OH H H h0101010i
6 Casticin – OMe @O OH OMe OMe H H OH OMe H H h0001000i
7 Catechin H OH 2H OH H OH H H OH OH H H h0110011i
8 Chrysin – H @O OH H OH H H H H H H h0001000i
9 Chrysoplenetin – OMe @O OH OMe OMe H H OMe OH H H h0001000i
10 Chrysosplenol-C – OMe @O OH OH OMe H H OMe OH H H h0001000i
11 Chrysosplenol-D – OMe @O OH OMe OMe H H OH OH H H h0101010i
12 Chysoeriol – H @O OH H OH H H OMe OH H H h0001001i
13 Cirsilineol – H @O OH OMe OMe H H OMe OH H OH h0001000i
14 Cirsimaritin – H @O OH OMe OMe H H H OH H H h0001010i
15 Cyanidin – OH H OH H OH H H OH OH H H h0110111i
16 Cyanidin-3-rutinoside – ORut H OH H OH H H OH OH H H h0100111i
17 Cyanin – OGlc H OH H OH H H OH OH H H h0100111i
18 Cyranoside – H @O OH H OGlu H H OH OH H H h0101010i
19 Delphinidin – OH H OH H OH H H OH OH OH H h0110111i
20 Epicatechingallate H OGall, H 2H OH H OH H H OH OH H H h1100011i
21 Epigallocatechin H OH, H 2H OH H OH H H OH OH OH H h0110011i
22 Epigallocatechin-3-O-ferulate H OFer, H 2H OH H OH H H OH OH OH H h1100011i
23 Epigallocatechingallate H OGall, H 2H OH H OH H H OH OH OH H h1100011i
24 Eupatorin – H @O OH OMe OMe H H OH OMe H H h0001000i
25 Fisetin – OH @O H H OH H H OH OH H H h0001011i
26 Galangin – OH @O OH H OH H H H H H H h0011011i
27 Gossypentin-3.8-dimethyl ether – OMe @O OH H OH OH H OH OMe H H h0001001i
28 Hesperetin H OH, H @O OH H OH H H OH OMe H H h0010001i
29 Hesperidin H 2H @O OH H OGlucRh H H OH OMe H H h0000000i
30 Hispiduline – H @O OH OMe OH H H H OH H H h0001010i
31 Ideain – OGal H OH H OH H H OH OH H H h0100111i
32 Isokaemferide – OMe @O OH H OH H H H OH H H h0001011i
33 Isorhamnetin – OH @O OH H OH H H OMe OH H H h0011001i
34 Kaemferol-6-methoxy-3-O-b-D-glucoside – OGlu @O OH OMe OH H H H OH H H h0001010i
35 Kaempferol – OH @O OH H OH H H H OH H H h0011011i
36 Keracyanin – ORut H OH H OH H H OH OH H H h0100111i
37 Keyakinine – OH @O OH ObGlcp OMe H H H OH H H h0011010i
38 Laricitrin – OH @O OH H OH H H OH OH OMe H h0111001i
39 Luteolin – H @O OH H OH H H OH OH H H h0101011i
40 Luteolin-30 -7-glucoside – H @O OH H OGlu H H OGlu OH H H h0001000i
41 Luteolin-40 -glucoside – H @O OH H OH H H OH OGlu H H h0001001i
42 Luteolin-7-methylether – H @O OH H OMe H H OH OH H H h0101010i
43 Malvidin – OH H OH H OH H H OMe OH OMe H h0010101i
44 Mearnsetin-7-glucoside – OH @O OH H OGlu H H OH OMe OH H h0011000i
45 Mikanin – OH @O OH OMe OMe H H H OMe H H h0011000i
46 Morin – OH @O OH H OH H H OH H OH H h0011011i
47 Myricetin – OH @O OH H OH H H OH OH OH H h0111011i
48 Naringenin H 2H @O OH H OH H H H OH H H h0000011i
49 Prunin H 2H @O OH H OGluc H H H OH H H h0000010i
50 Naringin H 2H @O OH H OR* H H H OH H H h0000010i
51 Narirutin H 2H @O OH H ORut H H H OH H H h0000011i
52 Oein – OGly H OH H OH H H OMe OH OMe H h0000101i
53 Padmatine H OH, H @O OH H OMe H H OH OH H H h0110010i
54 Pelargonidin – OH H OH H OH H H H OH H H h0010111i
55 Peonidin – OH H OH H OH H H OMe OH H H h0010101i
56 Quercetagenin-30 ,40 -dimethyl ether – OMe @O OH OH OH H H OH OMe H H h0001001i
57 Quercetin – OH @O OH H OH H H OH OH H H h0111011i
58 Rahmnetin – OH @O OH H OMe H H OH OH H H h0111010i
59 Ramnocitrine – OH @O OH H OMe H H H OH H H h0011010i
60 Retusin – OMe @O OH H OMe H H OMe OMe H H h0001000i
61 Rutin – ORut @O OH H OH H H OH OH H H h0101011i
62 Syringetin – OH @O OH H OH H H OMe OH OMe H h0011001i
63 Tamarixetin – OH @O OH H OH H H OH OMe H H h0011001i
64 Taxifolin H OH, H @O OH H OH H H OH OH H H h0110011i
65 3,5-Dihydroxy-30 ,40 ,60 ,70 -tetramethoxyflavone – OH @O OH OMe OMe H H OMe OMe H H h0011000i
66 20 ,30 -Dihydroxyflavone – H @O H H H H OH OH H H H h0101011i
67 3,6,40 -Trihydroxyflavone – OH @O H OH H H H H OH H H h0011011i
68 3,7,40 Trihydroxyflavone – OH @O H H OH H H H OH H H h0001011i
69 7,8,20 -Trihydroxyflavone – H @O H H OH OH OH H H H H h0001011i
70 7,8,30 -Trihydroxyflavone – H @O H H OH OH H OH H H H h0001011i
71 5,6,7-Trihydroxyflavone – H @O OH OH OH H H H H H H h0001011i
72 5,7,30 ,40 -Tetrahydroxy-2-methoxyflavone – OMe @O OH H OH H H OH OH H H h0101011i
73 5,30 ,40 -Trihydroxyflavone – H @O OH H H H H OH OH H H h0101010i
74 3,30 ,40 -Trihydroxyflavone – OH @O H H H H H OH OH H H h0101011i
*
R = [2-O-(6-Deoxy-a-L-mannopyranosyl)-b-D-glucopyranosyl].
 G. Castellano et al. / Phytochemistry 93 (2013) 182–191 185

Table 2
Classification level, number of classes and information entropy for the flavonoids.

Classification level b Number of classes Entropy h

1.00 74 2555.95
0.99 25 301.12
0.98 17 138.82
0.97 15 106.55
0.96 11 57.39
0.95 10 47.30
0.94 9 38.68
0.91 8 29.04
0.89 6 15.28
0.88 5 11.20
0.76 4 5.76
0.75 3 3.77
0.50 2 1.47
0.16 1 0.05

Class 1

Class 2

Fig. 2. Simplified partial correlation diagram: High (red) and medium (orange) Class 3
correlations. (For interpretation of the references to color in this figure legend, the
reader is referred to the web version of this article.) Class 4

14 Class 5
correlations show a similar vector property. With the equipartition
2 17
conjecture, the partial-correlations matrix contains 1300 high, Class 6
1188 medium, 81 low and 132 zero partial correlations. With the 10 5
12a 16
equipartition conjecture, the intercorrelations are calculated for Class 7
the partial-correlation diagram. We have selected a set of 20 points 12b
13
with great antioxidant activity, resulting in 106 high and 84 med- Class 8
1a 11 9a 6
ium partial correlations (Fig. 2). Flavonoids with greatest activity in 0 8a
F2 2a 15a 4a Class 9
9b
1b 8b 4b
15b Class 10
2b 3b
Class 11
3a
-2
Class 12

7 Class 13

-1 0 1 2 Class 14
F1
Class 15

Class 16

Class 17

Fig. 4. Principal component analysis F2 vs. F1 scores plot for flavonoids.

the same class are linked by red lines, representing high partial
correlations.
The grouping rule in the case with equal weights ak = 0.5, for the
classification level b = 0.98 allows 17 classes (grouped from class 1
to class 17) between parentheses:

Fig. 3. Dendrogram for the studied flavonoids.
C b ¼ð1; 3; 6; 8; 9; 10; 12; 13; 24; 27; 40; 41; 56; 60Þ;
ð2; 4; 14; 25; 30; 32; 34; 68; 69; 70; 71Þ;
ð5; 11; 18; 39; 42; 61; 66; 72; 73; 74Þ; ð7; 21; 53; 64Þ;
ð15; 19Þ; ð16; 17; 31; 36Þ; ð20; 22; 23Þ; ð47; 57; 58Þ;
ð26; 35; 37; 46; 59; 67Þ; ð28Þ; ð29Þ; ð33; 44; 45; 62; 63; 65Þ;
ð38Þ; ð43; 55Þ; ð48; 49; 50; 51Þ; ð52Þ; ð54Þ
The classes are obtained with the associated entropy
h(Rb) = 138.82. The dendrogram (Fig. 3) or binary tree (Jarvis and
Patrick, 1973; Page, 2000; Tryon, 1939) matching to hi1, i2, i3, i4,
186 G. Castellano et al. / Phytochemistry 93 (2013) 182–191

Table 3
Classification results of the studied flavonoids. Generally, groups of compounds with similar structures follow similar antioxidant activity.

g000 g001 g010 g011 g110

*
p000X 29 Flavanone 28 Flavanone
p001X* 48, 49, 50, 51 Flavanones 7, 21 Cathechins 20, 22, 23
53, 64 Flavanones Flavanols
p0101 52 Anthocyanin 43, 55 Anthocyanidins
p0111 54 Anthocyanidin 16, 17, 31, 36 Anthocyanins 15, 19
Anthocyanidins
p100X* 1, 3, 6, 8, 9, 10,12, 13, 24, 27, 40, 41, 56 33, 44, 45,46, 62, 63, 65 38 Flavonol
Flavones Flavonols
p101X* 2, 4, 14, 25, 30, 32, 34, 68, 69, 70, 71 26, 35, 37, 46, 59, 67 5, 11, 18, 39, 42, 61, 66, 72, 73, 74 47, 57, 58
Flavones Flavonols Flavonols Flavonols
*
X = 0 or 1.

Table 4
Antioxidant activity of flavonoid compounds.

⁄
X = 0 or 1.
a
Taken from Yang et al. (2001).
b
Taken from Anouar et al. (2009).
c
Taken from Zhang et al. (2007).

i5, i6, i7i and Cb is calculated (Shaw, 2003). In particular, the flavo-
noids with the greatest activity (20, 22, 23) are grouped into the
same class, corresponding to esters of cinnamic and benzoic acids
in 3-position on ring C. The dendrogram separates the same 17
classes in agreement with the partial correlation diagram (Fig. 2).
A comparative analysis of the set above from 74 classes (each
compound in its own class) to one class (containing all com-
pounds) is summarized in Table 2.
In the principal component analysis (PCA) F2–F1 scores plot
(Fig. 4), flavonoids with the same vector property appear superim-
posed. The same 17 classes of flavonoids are clearly distinguished
in concordance with the reference clustering: class 1 (compounds
1, 3, 6, 8, 9, 10, 12, 13, 24, 27, 40, 41, 56, 60, F1 < F2, left). . . class 7
(compounds 20, 22, 23, F1  F2, right bottom), etc. The classification
is in qualitative agreement with the partial-correlation diagram
and dendrogram (Figs. 2 and 3).
3. Results and discussion
The recommended format for the periodic table (PT) of flavo-
noids, cf. Table 3, shows that they are classified first by i1, then
by i2, i3, all standing for the group, and, finally, by i4, i5, i6, i7, denot-
ing the period. Periods of five units are assumed, which correspond
to groups; e.g., group g000 stands for hi1, i2, i3i = h000i, viz.
h000000Xi (hesperidin); h000001Xi (naringenin, etc.); h0000101i
(oein), etc., where X = 0 or 1.The flavonoids in the same column ap-
pear close in the partial-correlation diagram, dendrogram and PCA
(Figs. 2–4).
Table 4 includes data of antioxidant power taken from the liter-
ature (Apak et al., 2007; Firuzi et al., 2005; Pietta, 2000; Rice-Evans
et al., 1996). As the data have been obtained from a great variety of
experimental protocols, it has been very difficult for us to obtain a
unified set of reference data. Antioxidant character increases
 G. Castellano et al. / Phytochemistry 93 (2013) 182–191 187

across the periods, i.e. for period p001X: g000 < g011 < g110. In
most cases, especially for the groups in the right side, it also in-
creases, in general, when descending in a group, i.e. for g011:
p001X < p0111 < p101X. This order is the same as that found in
Table 3.
Our results above are also in agreement with Patra and Chua
(2011), in which they predicted 10 potent compounds with strong
antioxidant activity. These compounds are present in our period
p101X, which lies in the bottomof the table and, according to our
results, they coincide with those that show the greatest antioxi-
dant activity.
Fig. 5 exhibits the variation of property P of vector
hi1, i2, i3, i4, i5, i6, i7i, expressed in the decimal system
P = 106i1 + 105i2 + 104i3 + 103i4 + 102i5 + 10i6 + i7 as a function of
the structural parameters {i1, i2, i3, i4, i5, i6, i7} for the flavonoids.
In accordance with the previously assigned order, parameter
i1 shows greater variation than i2 than i3, etc. The antioxidant activ-
ity was not used in the development of the PT (Table 3) and serves
to validate it. The results agree with a PT of properties with vertical
groups defined by {i1, i2, i3} and horizontal periods described by
{i4, i5, i6, i7}.
The variation of property P of vector hi1, i2, i3, i4, i5, i6, i7i in base
10 as a function of the number of the group in PT for the flavonoids,
cf. Fig. 6, reveals minima corresponding to flavonoids with hi1, i2, i3,
i4, i5, i6, i7i close to h000i (group g000) and maxima near h110i
(group g110). Periods p000X, p001X, p0101, p0111, p100X,
p101X represent rows 1, 2, 3, 4, 5 and 6 in Table 3, respectively.
The corresponding function P (i1, i2, i3, i4, i5, i6, i7) reveals a series
of periods (periodic waves) clearly limited by maxima and minima,
which suggest a periodic behavior that recalls the form of a trigo-
nometric function. For h i1, i2, i3, i4, i5, i5, i6, i7 i a minimum is clearly
shown for most periods. Some points appear superimposed, espe-
cially the p101X line is on p0111 and p100X. The distance in hi1, i2,
i3, i4, i5, i6, i7i units between each pair of consecutive minima is five,
which coincides with the flavonoid sets belonging to the same
group in PT and in the successive periods. The minima occupy anal-
ogous positions in the curve and are in phase. The representative
points in phase should correspond to the elements of the same
group in PT. For hi1, i2, i3, i4, i5, i6, i7i minima there is coherence be-
tween the two representations; however, this coherence is not
complete. Only p101X is almost complete. It should be remarked
that the results for all periods present similar trends.
An empirical function P(p) reproduces the different hi1, i2, i3, i4,
i5, i6, i7i values. A minimum of P(p) has meaning only if it is com-
pared with the former P(p  1) and later P(p + 1) points, needing
to fulfill:
Pmin ðpÞ < Pðp  1Þ
Pmin ðpÞ < Pðp þ 1Þ ð6Þ
Order relations (6) should repeat at determined intervals equal
to the period size and are equivalent to:

i1 1000000
1000000
Vector property <i 1 , i2 , i3 , i4 , i5 , i6 , i7 >

i2 p000X

i3 p001X

p0101
i4 0
Z(p)
500000 p0111
i5
p100X
i6
p101X
i7
-1000000

0
0 1 2 4
Structural parameter Group number

Fig. 5. Variation of vector property of flavonoids in base 10 vs. counts {i1, i2, i3, i4, i5, Fig. 7. Variation of Z(p) = P(p + 1)  P(p) vs. group number for flavonoids.
i6, i7}.

100

1000000 p000X

p000X
p001X
p001X
Vector property

p0101
p0101 R(p) 50
p0111
500000 p0111
p100X
p100X
p101X
p101X

2 4
2 4 Group number
Group number
Fig. 8. Fragment of the variation of R(p) = P(p + 1)/P(p) vs. group number for
Fig. 6. Variation of vector property of flavonoids in base 10 vs. group number. flavonoid compounds.
Table 5

188
Structures according to classification of the studied flavonoids.

g000 g001 g010 g011 g110

p000X*

p001X*

G. Castellano et al. / Phytochemistry 93 (2013) 182–191

p0101

p0111
Table 5 (continued)

g000 g001 g010 g011 g110

p100X*

G. Castellano et al. / Phytochemistry 93 (2013) 182–191

p101X*

*
X = 0 or 1.

189
190 G. Castellano et al. / Phytochemistry 93 (2013) 182–191
Pmin ðpÞ  Pðp  1Þ < 0
Pðp þ 1Þ  Pmin ðpÞ > 0
As relations (7) are valid only for minima, more general others
are desired for all the values of p. The differences Z(p) are calcu-
lated assigning each of their values to flavonoid compound p:
ZðpÞ ¼ Pðp þ 1Þ  PðpÞ
Instead of Z(p) the R(p) = P(p + 1)/P(p) values can be taken
assigning them to flavonoid compound p. If periodic law (PL) were
general, the elements in the same group in analogous positions in
the different periods would satisfy:
either ZðpÞ > 0 or ZðpÞ < 0
either RðpÞ > 1 or RðpÞ < 1
Fig. 7 shows variation of Z(p) vs. group number. The results, in
general, are as expected, Z(p) > 0 for all periods according to Eq.
(9) and the periodic law. For group 5, the result Z(p) < 0 should
be taken with care, because Z(p) is calculated from the following
position in the periodic table in the subsequent period.
Moreover, Fig. 8 shows the change of R(p) vs. group number,
pointing up that R(p) > 1 is in concordance with Eq. (10) and the
periodic law.
Table 5 shows the structures clustered into groups and periods.
When comparing the results in the present classification with the
experimental ones reported in the literature, it is observed in all
the groups that the nature of the substituents on ring A (OH or
OR) barely influences the antioxidant activity, so that they have
not been drawn.
All the flavonones (g000, g001 and g011 in p000X and p001X)
and catechins (g011 in p001X) above are the least antioxidant
flavonoids as there is only one (C@C) double bond in rig C, so that
the radical that can be formed presents a structure lesser stabilized
by resonance and lesser planar. On passing to period p0101 the
presence of these double bonds stabilizes the structure by reso-
nance, increasing planarity and so augmenting the antioxidant
activity.
In all periods above, it can be seen that when passing from
(g001) to (g110) the presence of OH in ortho position on ring B is
the most important factor in the antioxidant character because of
the importance that has this structural fragment to scavenge free
radicals, as it is widely described in the technical literature.
In addition, in all the periods above it is observed that on pass-
ing from (g000) to (g001) and from (g010) to (g011) the presence
in R3 on ring C of an OH group increases the antioxidant activity
against the presence of an ether OR group in the same position.
It should be remarked that in this position there is an ester group
of gallic or ferulic acid (g110), the antioxidant character increases
considerably by the resonant contribution from these esters.
When descending from period p0111 to periods p100X to
p101X the antioxidant activity above increases. This means that
flavones and flavonols are more antioxidant than anthocyanins
and anthocyanidins, as they are stabilized by resonance because
of the presence of a keto group in position 4 on ring C conjugated
with the double bond and that hydrogen bonds with the OH in R5
on ring A. This effect is even greater in these last periods when
passing from (g000) to (g001) because an OH is introduced in R3
on ring C, which also adds a hydrogen bond with the keto function.
4. Conclusions
This report presents results of classification of the 74 studied
flavonoid compounds. The analysis includes such chemical com-
pounds that fit the following general scheme: either cinnamic or
benzoic ester group at 3-position on ring C, ortho-dihydroxy (cate-
chol) structure on the B ring, the presence of two hydroxyl groups
ð7Þ at positions 3 and 5, the presence of a 2,3-double bond in conjuga-
tion with a 4-keto function on the B ring, a 2,3-double bond and
4,5-double bond, the absence of alkoxyl and glycoxyl ester groups
on ring B, and the lack of alkoxyl and glycoxyl ester groups on ring
A.
Generally, the groups of compounds with similar structure fol-
ð8Þ
low the same trend. The antioxidant character increases through
the periods. In most cases it also augments when descending in a
group.
In this work we have given priority to structural requirements
for efficient antioxidant potency.
Information entropy and principal component analysis allow
ð9Þ classifying the flavonoid compounds and agree.
ð10Þ The obtained periodic table assigns different groups and periods
to distinguished types of flavonoids, according to the kinds of the
classical classification, e.g. flavanones, etc. It should be remarked
that flavanols, esterified in position 3 with cinnamic acid deriva-
tives (in the group g110 at the right side of the periodic table),
show the greatest antioxidant activity as confirmed by experimen-
tal data in the literature. Furthermore, the second group from the
right (g011) contains cathechin, flavanone, anthocyanidin and fla-
vonol compounds with at least two hydroxyl groups in 30 and 40
positions of ring B.
In this work, we conclude that the antioxidant activity in flavo-
noid compounds depends on specific fragments present in the
structure, in the following order of importance:
1. Presence of gallic or ferulic ester in R3 on ring C that causes an
added stability by resonance.
2. Presence of OH groups in ortho on ring B.
3. Presence of a keto function conjugated with a double bond in
ring C that hydrogen bonds with OH in R3 on ring C and R5 on
ring A.
4. Presence of conjugated double bonds (C@C with C@C or C@O) in
ring C.
5. Presence of OH in R3 on ring A.
6. The presence of ether groups (OR) on rings C and B decreases
the antioxidant character, as they are electron donor groups
that destabilize the radicals and may reflect steric effects that
perturb planarity.
7. Apparently the OR and OH on ring A hardly influence the anti-
oxidant activity.
Acknowledgment
This work was supported by the Spanish Ministerio de Ciencia e
Innovación (Project No. BFU2010-19118).
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.phytochem.2013.
03.024.
References
Anouar, E., Kosinová, P., Kozlowski, D., Mokrini, R., Duroux, J.L., Trouillas, P., 2009.
New aspects of the antioxidant properties of phenolic acids: a combined
theoretical and experimental approach. Phys. Chem. Phys. 11, 7659–7668.
Apak, R., Güalü, K., Demirata, B., Ózyürek, M., Celik, S.E., Bektasoglu, B., Berker, K.I.,
Ózyurt, D., 2007. Comparative evaluation of various total antioxidant capacity
assays applied to phenolic compounds with the CUPRAC assay. Molecules 12,
1496–1547.
Bors, W., Heller, W., Michel, C., Saran, M., 1990. Flavonoids as antioxidants:
determination of radical-scavenging efficiencies. Methods Enzymol. 186, 343–
355.
 G. Castellano et al. / Phytochemistry 93 (2013) 182–191
Brand-Williams, W., Cuvelier, M.E., Berset, C., 1995. Use of free radical method to
evaluate antioxidant activity. Food Sci. Technol. 28, 25–30.
Castellano, G., Tena, J., Torrens, F., 2012. Classification of phenolic compounds by
chemical structural indicators and its relation to antioxidant properties of
Posidonia Oceanica (L.) Delile. MATCH Commun. Math. Comput. Chem. 67, 231–
250.
Cao, G., Sofic, E., Prior, R.L., 1997. Antioxidant and pro-oxidant behavior of
flavonoids: structure–activity relationships. Free Radic. Biol. Med. 22, 749–760.
Croft, K.D., 1998. The chemistry and biological effects of flavonoids and phenolic
acids. Ann. N. Y. Acad. Sci. 854, 435–442.
Dewick, P.M., 2002. Medicinal Natural Products: A Biosynthetic Approach. Ed.
Wiley, New York.
Dumay, O., Costa, J., Desjobert, J., Pergent, G., 2004. Variations in the concentration
of phenolic compounds in the seagrass Posidonia oceanica under conditions of
competition. Phytochemistry 65, 3211–3220.
Firuzi, O., Lacanna, A., Petrucci, R., Marrosu, G., Saso, L., 2005. Evaluation of the
antioxidant activity of flavonoids by ‘‘ferric reducing antioxidant power’’ assay
and cyclic voltammetry. Biochim. Biophys. Acta 1721, 174–184.
Fukumoto, L.R., Mazza, G., 2000. Assessing antioxidant and prooxidant activities of
phenolic compounds. J. Agric. Food Chem. 48, 3597–3604.
Halliwell, B., Aeschbach, R., Loliger, J., Aruoma, O.I., 1995. The characterization of
antioxidants. Food Chem. Toxicol. 33, 601–617.
Iordache, O., Corriou, J.P., Garrido-Sánchez, L., Fonteix, C., Tondeur, D., 1993. Neural
network frames. Application to biochemical kinetic diagnosis. Comput. Chem.
Eng. 17, 1101–1113.
Jarvis, R.A., Patrick, E.A., 1973. Clustering using a similarity measure based on
shared nearest neighbors. IEEE Trans. Comput. 22, 1025–1034.
Kadoma, Y., Fujisawa, S., 2008. A comparative study of the radical-scavening activity
of the phenolcarboxylic acids caffeic acid, p-coumaric acid, chlorogenic acid and
ferulic acid, with or without 2-mercaptoethanol, a thiol, using the induction
period method. Molecules 13, 2488–2499.
Kaufmann, A., 1975. Introduction à la Théorie des Sous-ensembles Flous, vol. 3. Ed.
Masson, Paris.
191
Khadem, S., Marles, R.J., 2010. Monocyclic phenolic acids; hydroxy- and
polyhydroxybenzoic acids: occurrence and recent bioactivity studies.
Molecules 15, 7985–8005.
Kullback, S., 1959. Information Theory and Statistics. Ed. Wiley, New York.
Lacikova, L., Jancova, M., Muselik, J., Masterova, I., Grancai, D., Fickova, M., 2009.
Antiproliferative, cytotoxic, antioxidant activity and polyphenols contents in
leaves of four Staphylea L. species. Molecules 14, 3259–3267.
Omata, Y., Saito, Y., Yoshida, Y., Niki, E., 2008. Simple assessment of radical
scavenging capacity of beverages. J. Agric. Food Chem. 56, 3386–3390.
Page, R.D.M., 2000. Program TreeView. Ed. University of Glasgow.
Patra, J.C., Chua, B.H., 2011. Artificial neural network-based drug design for diabetes
mellitus using flavonoids. J. Comput. Chem. 32, 555–567.
Pietta, P., 2000. Flavonoids as antioxidants. J. Nat. Prod. 63, 1035–1042.
Rice-Evans, C.A., Miller, N.J., Paganga, G., 1996. Structure–antioxidant activity
relationships of flavonoids and phenolic acids. Free Radic. Biol. Med. 20, 933–
956.
Shaw, P.J.A., 2003. Multivariate Statistics for the Environmental Sciences. Ed.
Hodder-Arnold, New York.
Simic, A., Manojlovic, D., Segan, D., Todorovic, M., 2007. Electrochemical behavior
and antioxidant and prooxidant activity of natural phenolics. Molecules 12,
2327–2340.
Tryon, R.C., 1939. A multivariate analysis of the risk of coronary heart disease in
Framingham. J. Chronic Dis. 20, 511–524.
White, H., 1989. Neural network learning and statistics. AI Expert 4 (12), 48–52.
Yang, C.S., Landau, J.M., Huang, M., Newmark, H.L., 2001. Inhibition of
carcinogenesis by dietary polyphenolic compounds. Annu. Rev. Nutr. 21, 381–
406.
Yi, L., Jin, X., Chen, Ch., Fu, Y., Zhang, T., Chang, H., Zhou, Y., Zhu, J., Zhang, Q., Mi, M.,
2001. Chemical structures of 4-oxo-flavonoids in relation to inhibition of
oxidized low-density lipoprotein (LDL)-induced vascular endothelial
dysfunction. Int. J. Mol. Sci. 12, 5471–5489.
Zhang, J., Stanley, R.A., Melton, L.D., Skinner, M.A., 2007. Inhibition of lipid oxidation
by phenolic antioxidants in relation to their physicochemical properties.
Pharmacol. Online 12, 1496–1547.