Professional Documents
Culture Documents
Received 21 November 2003; received in revised form 21 January 2004; accepted 21 January 2004
Abstract
A review of the use of the Doehlert matrix as a chemometric tool for the optimization of methods in analytical chemistry and other sciences
is presented. The theoretical principles of Doehlert designs are described, including the coded values for the use of this matrix involving two,
three, four and five variables. The advantages of this matrix in comparison with other response surface designs, such as central composite and
Box–Behnken, designs are discussed. Finally, 57 references concerning the application of Doehlert matrices in the optimization of procedures
involving spectroanalytical, electroanalytical and chromatographic techniques are considered.
© 2004 Elsevier B.V. All rights reserved.
1. Introduction relatively low cost, full factorial designs are very useful for
preliminary studies or in the initial steps of an optimiza-
As complexity, variety, growing importance of quality and tion, while fractional designs are almost mandatory when
omnipresent uncertainty mark today’s chemistry, a statistical the problem involves a large number of factors [3]. How-
approach to experimental design is almost inevitable. There- ever, since only two levels are used, the models that can be
fore, the use of multivariate experimental design techniques fit to these designs are somewhat restricted. Consequently,
is becoming increasingly widespread in analytical chemistry if a more sophisticated model is required, as for the loca-
and other sciences. Multivariate designs, which allow the si- tion of an optimum set of experimental conditions, then one
multaneous study of several control variables, are faster to must resort to designs for second-order models (response
implement and more cost-effective than traditional univari- surface designs), which employ more than two factor levels
ate approaches [1,2]. Several experimental design models to allow fitting of a full quadratic polynomial. In analytical
exist that reduce the number of experiments and that can be chemistry, the response surface designs most used are sym-
used in different cases. Thus, if it is desired to detect in- metrical and describe a spherical experimental domain.
fluential factors, experimental designs for first-order models A central composite design (CCD) [4] combines a
(factorial designs or Plackett–Burman designs) can be used. two-level full or fractional factorial design with additional
On the other hand, to approximate a response function or to points (star points) and at least one point at the center of
optimize a process, experimental designs for second-order the experimental region, selected to obtain properties such
models should be used. as rotatability or orthogonality, in order to fit quadratic
The most popular first-order design is the two-level full polynomials. For fitting quadratic response models, the
(or fractional) factorial, in which every factor is experimen- CCD is a better alternative to the full factorial three-level
tally studied at only two levels. Due to their simplicity and design because its performance is comparable at a lower
cost. Therefore, it has been the most accepted experimen-
tal design for second-order models. The total number of
∗ Corresponding author. Tel.: +55-71-2355166; fax: +55-71-2355166. design points needed (N) is determined by the formula
E-mail address: slcf@ufba.br (S.L.C. Ferreira). N = 2k + 2k + C0 , where k is the number of variables and
0039-9140/$ – see front matter © 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.talanta.2004.01.015
1062 S.L.C. Ferreira et al. / Talanta 63 (2004) 1061–1067
Table 1 Table 3
Comparison of efficiency of central composite design (CCD), Box– Doehlert matrix for three-variables
Behnken design (BBD) and Doehlert design (DM)
Experimental variables
Variables Number of Number of Efficiency (p/f)
A B C
(K) coefficients (p) experiments (f)
CCD DM BBD CCD DM BBD 1 0 0 0
2 1 0 0
2 6 9 7 – 0.67 0.86 – 3 0.5 0.866 0
3 10 15 13 13 0.67 0.77 0.77 4 0.5 0.289 0.817
4 15 25 21 25 0.60 0.71 0.60 5 −1 0 0
5 21 43 31 41 0.49 0.68 0.61 6 −0.5 −0.866 0
6 28 77 43 61 0.36 0.65 0.46 7 −0.5 −0.289 −0.817
7 36 143 57 85 0.25 0.63 0.42 8 0.5 −0.866 0
8 45 273 73 113 0.16 0.62 0.40 9 0.5 −0.289 −0.817
10 −0.5 0.866 0
11 0 0.577 −0.817
C0 is the number of center points. In analytical chemistry, 12 −0.5 0.289 0.817
this design has been widely used. 13 0 −0.577 0.817
Box–Behnken designs (BBD) [5] are rotatable second-
order designs based on three-level incomplete factorial de-
signs. The special arrangement of the BBD levels allows the and the different applications of these experimental designs
number of design points to increase at the same rate as the in analytical chemistry and other sciences.
number of polynomial coefficients. For three factors, for ex-
ample, the design can be constructed as three blocks of four
experiments consisting of a full two-factor factorial design 2. Doehlert matrices
with the level of the third factor set at zero. The number of
experimental points (N) is defined by the expression N = The Doehlert design describes a spherical experimental
2k(k − 1) + C0 , where k is the number of variables and C0 domain and it stresses uniformity in space filling. Although
is the number of center points. this matrix is neither orthogonal nor rotatable, it does not
An alternative and very useful experimental design for significantly diverge from the required quality for effective
second-order models is the uniform shell design proposed use [7]. For two variables, the Doehlert design consists of
by Doehlert in 1970 [6]. Doehlert designs are easily applied one central point and six points forming a regular hexagon,
to optimize variables [7,8] and offer advantages in relation and therefore situated on a circle. In three dimensions it can
to central composite and Box–Behnken designs. They need be viewed in different ways, depending on the geometric
fewer experiments, which are more efficient and can move
through the experimental domain. Despite these attractive
Table 4
features, however, it took some time until researchers started Doehlert matrix for four-variables
to take notice of Doehlert designs.
Experimental variables
In analytical chemistry, the first application of the
Doehlert matrix involved the optimization of a separation A B C D
process using HPLC [9]. Since then, several papers have 1 0 0 0 0
appeared involving determinations by spectrometric, chro- 2 1 0 0 0
matographic and electroanalytical methods. 3 0.5 0.866 0 0
4 0.5 0.289 0.817 0
This paper offers a critical discussion of the different
5 0.5 0.289 0.204 0.791
chemometric approaches developed for the optimization of 6 −1 0 0 0
chemical and instrumental variables using Doehlert matrices 7 −0.5 −0.866 0 0
8 −0.5 −0.289 −0.817 0
Table 2 9 −0.5 −0.289 −0.204 −0.791
Doehlert matrix for two-variables 10 0.5 −0.866 0 0
11 0.5 −0.289 −0.817 0
Experimental variables 12 0.5 −0.289 −0.204 −0.791
13 −0.5 0.866 0 0
A B
14 0 0.577 −0.817 0
1 0 0 15 0 0.577 −0.204 −0.791
2 1 0 16 −0.5 0.289 0.817 0
3 0.5 0.866 17 0 −0.577 0.817 0
4 −1 0 18 0 0 0.613 −0.791
5 −0.5 −0.866 19 −0.5 0.289 0.204 0.791
6 0.5 −0.866 20 0 −0.577 0.204 0.791
7 −0.5 0.866 21 0 0 −0.613 0.791
S.L.C. Ferreira et al. / Talanta 63 (2004) 1061–1067 1063
ables and simultaneously the minimum response for the lev- where R is the experimental response to be optimized, a is
els of other variables of the analytical system studied. the constant term, b, c and d are coefficients of the linear
terms, e, f and g are coefficients of the quadratic terms and
3.2. Canonical analysis h, i, and j are coefficients of interaction between the three
factors.
Canonical analysis [3] is a mathematical treatment which If the quadratic function only shows one stationary point
consists of changing the origin of the plot from its original (A0 , B0 , C0 ), four situations are possible:
co-ordinates to the stationary point and rotating the axes
until they correspond to the principal axes of the contours. 1. There is not any information: ∆2 = 0.
Using this new co-ordinate system, the second-order model 2. Relative maximum: ∆1 < 0; ∆2 > 0; ∆3 < 0.
equations are simplified and its geometrical nature becomes 3. Relative minimum: ∆1 > 0; ∆2 > 0; ∆3 > 0.
apparent. 4. Saddle point: none of the above situations applies.
The algebraic signs of the canonical equation coefficients
provide an idea about the nature of its stationary point. If
where ∆3 is the Hessian determinant of the function H (A, B,
the values are all negative it is a maximum; if all positive, it
C), ∆2 and ∆1 are calculated using the following equations:
is a minimum, and if the signs are mixed it is a saddle point.
Table 6
Papers involving spectroanalytical techniques optimized by Doehlert matrix
Analyte Optimization (factors number) Analytical technique Sample References
The Doehlert matrix has proved its usefulness on several The Doehlert matrix has also been applied to other fields
types of chromatography, where it was used to optimize the of science, due to its generality and ease of use. It was
separation of species and to increase the selectivity of the used to optimize process variables and in formulation and
methods. This matrix has been used in the systematic and quality control in the development of pharmaceutical prod-
1066 S.L.C. Ferreira et al. / Talanta 63 (2004) 1061–1067
ucts [47,48]. It was also applied to examine virus adsorption [11] J.M. Bosque-Sendra, M. Nechar, L. Cuadros Rodrı́guez, M.F. Molina,
[49,50]. Anal. Proc. 32 (1995) 375.
[12] M.J. Valderrama, Métodos matemáticos aplicados a las Ciencias
In materials science, the use of this matrix produced
Experimentales, Pirámide, Madrid, 1989, p. 290.
better yields in different processes, such as quality con- [13] T. Dagnac, A. Padro, R. Rubio, G. Rauret, Anal. Chim. Acta 364
trol of granulation in a high shear mixer [51], conversion (1998) 19.
coatings on stainless steel in nitric acid solution [52], [14] R.J. Cassella, O.D. de Sant’Ana, R.E. Santelli, Spectrochim. Acta,
adsorption of metallic ions onto fly ash [53], prepara- Part B 57 (2002) 1967.
tion of activated carbon from olive-waste cakes [54] and [15] M.V. Rebouças, S.L.C. Ferreira, B.B. Neto, J. Anal. At. Spectrom.
18 (2003) 1267.
coagulation–flocculation of raw water [55]. [16] M. Nechar, M.F. Molina, J.M. Bosque-Sendra, Anal. Chim. Acta
This optimization method was also applied to the exper- 382 (1999) 117.
imental conditions of enzymatic synthesis [56], in the pro- [17] S.L.C. Ferreira, M.A. Bezerra, W.N.L. dos Santos, B.B. Neto, Talanta
duction of pectate lyase by a recombinant Escherichia coli 61 (2003) 295.
[57], in the improvement of the production of Penicillium [18] D. Gazquez, M. Sanchez-Viñas, M.G. Bagur, G. Garcı́a, J. Anal. At.
Spectrom. 13 (1998) 105.
cyclopium lipase [58] and P. cyclopium partial acylglycerol
[19] A.M. Garcı́a-Campaña, F. Ales Barrero, A. Lupiañez González, M.
lipase [59]. Also, it was used to determine the effect of fac- Román Ceba, Anal. Chim. Acta 447 (2001) 219.
tors on the accumulation of trehalose and glycerol produc- [20] L. Gámiz Gracia, L. Cuadros Rodrı́guez, M. Román Ceba, Talanta
tion by Saccharomyces cerevisiae [60]. Other more eclectic 44 (1997) 75.
applications were the adsorption process of radioelements [21] M. Zougagh, A. Garcı́a de Torres, J.M. Cano Pavón, Anal. Lett. 36
on mixtures of minerals [61] and the investigation of the (2003) 1115.
[22] Z. Benzo, P. Araujo, A. Sierraalta, F. Ruette, Anal. Chem. 65 (1993)
thermal behavior of extractants used for reprocessing nu- 1107.
clear fuel [62]. [23] S.L.C. Ferreira, H.C. dos Santos, M.S. Fernandes, M.S. de Carvalho,
J. Anal. At. Spectrom. 17 (2002) 115.
[24] M.A. Bezerra, A.B. Conceição, S.L.C. Ferreira, Anal. Bioanal. Chem.
6. Conclusions 378 (2004) 798.
[25] S.L.C. Ferreira, W.N.L. dos Santos, M.A. Bezerra, V.A. Lemos, J.M.
Bosque-Sendra, Anal. Bioanal. Chem. 375 (2003) 443.
The Doehlert matrix is a good design for response sur- [26] M. Camino, M.G. Bagur, M. Sanchez-Viñas, D. Gazquez, R. Romero,
face methodology because it permits: (i) estimation of the J. Anal. At. Spectrom. 16 (2001) 638.
parameters of the quadratic model, (ii) building of sequen- [27] J.A.A. Amaro, S.L.C. Ferreira, J. Anal. At. Spectrom. 19 (2004)
tial designs, (iii) detection of lack of fit of the model and (iv) 246.
use of blocks. Also, the Doehlert design as a chemometric [28] S.L.C. Ferreira, A.S. Queiroz, M.S. Fernandes, H.C. dos Santos,
Spectrochim. Acta, Part B 57 (2002) 1939.
tool presents advantages over other, more used, response sur- [29] W.N.L. dos Santos, C.M.C. Santos, S.L.C. Ferreira, Microchem. J.
face designs, such as composite central and Box–Behnken. 75 (2003) 211.
Doehlert designs, despite their relatively scarce diffusion, [30] M.F. Molina, M. Nechar, J.M. Bosque-Sendra, Anal. Sci. 14 (1998)
are amply used in analytical chemistry and other sciences. 791.
Therefore, different applications of these designs in the op- [31] M. Zougagh, P. Cañada Rudner, A. Garcı́a de Torres, J.M. Cano
Pavón, J. Anal. At. Spectrom. 15 (2000) 1589.
timization of procedures involving several analytical tech-
[32] J.M. Bosque-Sendra, M. Nechar, M.F. Molina, Mikrochim. Acta 134
niques are presented in this paper. (2000) 43.
[33] S. Furlanetto, S. Orlandini, G. Aldini, R. Gotti, E. Dreassi, S. Pin-
zauti, Anal. Chim. Acta 413 (2000) 229.
References [34] S. Furlanetto, S. Pinzauti, P. Gratteri, E. La Porta, G. Calzeroni, J.
Pharm. Biomed. Anal. 15 (1997) 1585.
[1] D.C. Montgomery, Design and Analysis of Experiments, 4th ed., [35] S. Furlanetto, S. Pinzauti, E. La Porta, A. Chiarugi, P. Mura, S.
Wiley, New York, 1997. Orlandini, J. Pharm. Biomed. Anal. 17 (1998) 1015.
[2] B.B. Neto, I.S. Scarminio, R.E. Bruns, Como Fazer Experimentos, [36] P. Araujo, Trends Anal. Chem. 19 (2000) 524.
Unicamp, Campinas, 2001. [37] B. Bourguignon, F. Marcenac, H.R. Keller, P.F. Deaguiar, D.L. Mas-
[3] G.E.P. Box, W.G. Hunter, J.S. Hunter, Statistics for Experimenters, sart, J. Chromatogr. 628 (1993) 171.
Wiley, New York, 1978. [38] R. Fraile, V. Sanchez, J. High Resol. Chromatogr. 16 (1993) 169.
[4] G.E.P. Box, K.B. Wilson, J. R. Statist. Soc., B 13 (1951) 1. [39] I.P. Durand, S. Gautier, E. Robert, M.C. Guilhem, R. Phan-Tan-Luu,
[5] G.E.P. Box, D.W. Benhken, Technometrics 2 (1960) 195. J. High Resol. Chromatogr. 20 (1997) 289.
[6] D.H. Doehlert, Appl. Statist. 19 (1970) 231. [40] J. Krupèı́k, M. Greòa, I. Špánik, E. Benická, J. Hrouzek, I. Skaèáni,
[7] D.L. Massart, B.G.M. Vandeginste, L.M.C. Buydens, S. de Jong, P.J. P. Sandra, J. Chromatogr. A 779 (1997) 253.
Lewi, J. Smeyers-Verbeke, Handbook of Chemometrics and Quali- [41] L. Mateus, S. Cherkaoui, P. Christen, J.L. Veuthey, J. Chromatogr.
metrics, Part A, Elsevier, Amsterdam, 2003. A 829 (1998) 317.
[8] M. Nechar, M.F. Molina, L. Cuadros Rodrı́guez, J.M. Bosque-Sendra, [42] A. Navalón, A. Prieto, L. Araujo, J.L. Vı́lchez, J. Chromatogr. A
Anal. Chim. Acta 316 (1995) 185. 946 (2002) 239.
[9] Y. Hu, D.L. Massart, J. Chromatogr. A 485 (1989) 311. [43] C. Aguilar, A. Peñalver, E. Pocurull, J. Ferré, F. Borrull, R.M. Marcé,
[10] A.M. Garcı́a-Campaña, L. Cuadros Rodrı́guez, A. Lupiañez J. Chromatogr. A 844 (1999) 425.
González, F. Alés Barrero, M. Román Ceba, Anal. Chim. Acta 348 [44] L. Paugam, R. Ménard, J.P. Larue, D. Thouvenot, J. Chromatogr. A
(1997) 237. 864 (1999) 155.
S.L.C. Ferreira et al. / Talanta 63 (2004) 1061–1067 1067
[45] R. Romero, M.G. Bagur, M. Sanchez-Viñas, D. Gazquez, Chro- [54] A. Baçaoui, A. Yaacoubi, A. Dahbi, C. Bennouna, R. Phan-Tan-Luu,
matographia 51 (2000) 404. F.J. Maldonado-Hodar, J. Rivera-Utrilla, C. Moreno-Castilla, Carbon
[46] R. Romero, Ja. Jonsson, D. Gazquez, M.G. Bagur, M. Sanchez-Viñas, 39 (2001) 425.
J. Sep. Sci. 25 (2002) 584. [55] M. Franceschi, A. Girou, A.M. Carro-Diaz, M.T. Maurette, E. Puech-
[47] D. Voinovich, M. Moneghini, B. Perissutti, J. Filipovic-Grcic, I. Costes, Water Res. 36 (2002) 3561.
Grabnar, Int. J. Pharm. 203 (2000) 235. [56] A. Ismail, M. Linder, M. Ghoul, Enzyme Microb. Technol. 25 (1999)
[48] C. Sánchez-Lafuente, S. Furlanetto, M. Fernández-Arévalo, J. 208.
Alvarez-Fuentes, A.M. Rabasco, M.T. Faucci, S. Pinzauti, P. Mura, [57] C.G. Hounsa, J.M. Aubry, H.C. Dubourguier, J.P. Hornez, Appl.
Int. J. Pharm. 237 (2002) 107. Microbiol. Biotechnol. 45 (1996) 764.
[49] F. Quignon, A. Huyard, L. Schwartzbrod, F. Thomas, J. Virol. Meth. [58] G. Vanot, V. Deyris, M.-C. Guilhem, R. Phan-Tan-Luu, L.-C.
68 (1997) 33. Comeau, Appl. Microbiol. Biotechnol. 57 (2001) 342.
[50] F. Quignon, F. Thomas, C. Gantzer, A. Huyard, L. Schwartzbrod, [59] G. Vanot, D. Valérie, M.-C. Guilhem, R. Phan-Tan-Luu, L.-C.
Water Res. 32 (1998) 1222. Comeau, Appl. Microbiol. Biotechnol. 60 (2002) 417.
[51] D. Vojnovic, D. Chicco, H. El Zenary, Int. J. Pharm. 145 (1996) 203. [60] F. Carvalheiro, J.C. Roseiro, F.M. Gı́rio, Food Microbiol. 16 (1999)
[52] S.E. Hajjaji, M.-T. Maurette, E. Puech-Costes, A. Guenbour, A. Ben 543.
Bachir, L. Aries, Surf. Coat. Technol. 110 (1998) 40. [61] P. Jacquier, P. Meier, J. Ly, Appl. Geochem. 16 (2001) 85.
[53] P. Ricou-Hoeffer, I. Lecuyer, P. Le Cloirec, Water Res. 35 (2001) [62] T. Dagnac, J.-M. Guillot, P.L. Cloirec, J. Anal. Appl. Pyrolysis 37
965. (1996) 33.