See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/21951117

Application of modified Rosenbrock's method

for optimization of nutrient media used in
microorganism culturing

Article in Biotechnology and Bioengineering January 1976

DOI: 10.1002/bit.260171214 Source: PubMed

CITATIONS READS

11 21

3 authors, including:

Ales Prokop
Vanderbilt University
144 PUBLICATIONS 2,271 CITATIONS

SEE PROFILE

All content following this page was uploaded by Ales Prokop on 03 October 2014.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
BIOTECHNOLOGY AND BIOENGINEERdNG VOL. XVII (1975)

Application of Modi'ed Rosenbrock's Method for

Optimization of Nutrient Media Used in Microorganism
Culturing

INTRODUCTION
Economic exploitation of a microbial process requires an adjustment of the
optimal cultivation environment. Cultivation conditions include the source and
concentration of substrate and the mode of its supply, the source and concentra-
tion of nitrogen, phosphorus, vitamins, microelements and other components,
as well as the cult.ivation temperature, pH of the medium, dissolved oxygen
level, etc. The growth of an organism at constant temperature, aeration, and
pH can be described by a general model based on Monod-type kinetics coupled
with substrate (product) inhibition. The dependence of the specific growth rate
on the concentration of substrat,e is then given by the formula

1 ddtX
P = -- = Pmsx n"
1-1
(sj + ).(
-
uj
____
Sj/b, +1
)
where uj is the half saturation constant in the Monod-type term, bj is a constant
characterizing the inhibitory effect of a substrate of concentration Sj, and X is
biomass concentration. The inhibition term becomes significant mainly with
components (substrates) such as low-molecular fatty acids, ethanol, methanol,
etc. On the other hand it can be neglected when bj >> Sj (no inhibition).
Optimization of the nutrient medium in a multidimensional (parameter) space
in a surface defined by eq. (1) renders the experimental treatment impossible,
especially with several numbers of optimized parameters.
A classical optimization procedure consists of a reduction from a multiparam-
eter to a single-parameter optimization with the remaining parameters con-
stant, but the optimum is usually reached very slowly. Himmelblaul recom-
mends procedures for finding optimum in multidimensional systems. Gradient
methods, requiring the derivative of experimental variable, are represented by
the already classical Box-Wilson method.* In the case of multidimensional (n)
optimization, these methods become very laborious because, to obtain the deriva-
tive and consequently the direction of the steepest ascent, i t is necessary to carry
out a (2n)-factorial experiment. This was the main reason why the authors
turned their attention to other optimum search procedures which would not
require so many experiments. It was advantageous to apply direct search
+
methods requiring only (n 1) experiments during a single run. These include,
e.g., the methods of Hook-Jeeve~,~ R o ~ e n b r o c kPowell,6
,~ and the simplex design.6

OBJECTIVE FUNCTION
Along the selection of the right optimization method, the optimization criterion
represents the most important step. When considering economic factors, a
1833
@ 1975 by John Wiley & Sons, Inc.
1834 BIOTECHNOLOGY AND BIOENGINEERING VOL. XVII (1975)

simple function, related to profit, is obtained by combining nutrient costs and

profits (gain) related to product (biomass) costs produced during a given time

AX
F = gain = CB - - 2 ci si
At 1-1

If nutrient costs can be replaced by average costs, then a simpler objective func-
tion may be used:
AX
Fo = -- - LY Z Si (3)
At ,-I

where a is the ratio of average nutrient costs to product costs. The simplest
objective function is obtained in the case of negligible nutrient, costs as compared
with product costs, then

Fo = AX/At (4)

For the purpose of this work, viz. the optimization of nutrient medium for cultur-
ing yeasts on methanol, the optimization procedure of liosenbrock4 has been
modified for efficient planning of experiments.

ROSENBROCKS OPTIMIZATION PROCEDURE

Effective decision rules and procedure as applied to nutrient medium optimiea-

tion are as follows.
I ) Select an initial set of variables (parameters) Sl0,Szo,. S,O, magnitude of
individual changes (steps) k , (usually a half of S,O), and the lower and upper
limits of S,. The limits, especially the upper one, are introduced because of
costs of nutrients, especially growth factors.
2 ) Select a set of orthogonal vectors v, in such a way that this set is equivalent
to orthogonal vectors of the unit base in n-dimensional space.
3) An auxiliary variable d, is set to zero.
4) Carry out one step of optimization: to the last vector of optimal nutrient
composition (S10,. . S,o) add a step multiplied by vector vj, i.e., h,v,, and calcu-
late the composition of the medium for the following experiments. Be careful
not to exceed the upper or lower limits. If the limits are exceeded, replace the
calculated values of coordinates S, by the limit value which was surpassed.
5) Carry out experiments for a new composition and a comparison set @lo, ..
S,o), and find a rate of production (growth), usually in several steps.
6) Calculate the objective function according to 2 ) , 3), or 4), find a t what
points an improvement is reached and replace S,o by this new value. Add the
value of h, to the value of assessory variable d,. Multiply step h, by 3. I n the
opposite case, i.e., if no improvement is noted on changing h,, multiply the value
of h, by -0.5 and use the original value of S,O.
7) If the change of the objective function is within the limits of experimental
error, stop the calculation and end the optimization.
8) Check whether there is a t least one unsuccessful step in any of the direc-
tions. If this is so, repeat the whole procedure with a new h, value beginning
with item 4).
 COMMUNICATIONS TO T H E EDITOR 1835

9) Transform coordinates with the aid of the Gram-Schmidt orthogondiza-

tion according to the following relations: calculate vectors wk

wk = 2 d; ~j ( k = 1, . . ., n)
j- k

new set of vectors 0;- is given by

Using these new vectors v j n , compute a new calculation (beginning with item 3)
until the optimum is reached.

RESULTS
The above method has been used for the optimization of nutrient, medium for
culturing Candida boidinii 11 Bh on methanol in shake flasksi.8 a t two- and five-
parameter levels (medium components) and with the objective function defined
by eq. (4). It was shown that the modified Ilosenbrock's method is much more
convenient than the classical Box-Wilson method. The value of optimization
crit.erion, i.e., the optical density reached after 24 hr, is very much the same
regardless of which method has been used (Tables I and 11). The results were
compared with those obtained with an originally developed medium set, up
according t,o literary dat.a (Table 11). The economic value of t.he optimized
media is obvious since the medium compositions found by either of the optimiza-
tion methods give the same results with minimum amounts of ingredients.
Tables I and I11 also yield the number of experiment,al sets and measured variants
for t,he two- and five-parameter optimization. For many numbers of parameters,
the difference becomes considerable.

TABLE I
Optimal Concentration of Vitamins for Growth of C. boidinii 1 1 Bh in
Shake Flasks a t 1% v/v Methanol (n = 2 ) .

Biotin Thiamine Number of Number of

concentration concentration experimental measured
Method (pg/liter) (mg/liter) sets variants

Box-Wilson 7.5 2.6 6 33

Rosenbrock 7.3 2.8 7 21

8 Empirical medium (Table 11) with yeast extract replaced by vitamins.

1836 BIOTECHNOLOGY AND BIOENGINEERING VOL. XVII (1975)

TABLE I1
Comparison of Optimal Nutrient Composition with the Original Empirical
Medium for Growth of C. boidinii 11 Bh in Shake Flasks a t 1% v/v Methanol
(n = 5)

OD at
Yeast MgSO, . ZnSO,. 540 nm
extract Nitrogen KHzPO, 7Hz0 7 HzO after
Method (mg/liter) (g/liter) (g/liter) (g/liter) (mg/liter) 24 hr

Box-Wilson 1006 0.73 0.86 0.20 14.45 0.198

Rosenbrock 100" 0.67 1.35 0.22 8.00 0.199
Empirical
medium 100 1.32 7.00 0.50 10.00 0.200

a Upper limit set a t this value.

TABLE I11
Comparison of Optimization Methods with Respect t,o the Amount of Labor.

Number of Number of measured

Method experimental sets variants

Box-Wilson 6 101
1/2 Box-Wilsonb 6 61
Rosenbrock 5 29

a Conditions as in Table I1 (n = 5 ) .
b Box-Wilson method with half-factorial design, applicable for n 2 3.

CONCLUSION

The Rosenbrock's procedure has been modified for optimization of nutrient

medium composition and has been found to be less tedious than the Box-Wilson
method, especially for larger numbers of optimized parameters. Its merits
are particularly obvious with multiparameter optimization where the gradient
method, so far the only one employed in microbiology from a variety of opti-
mization methods (e.g., refs. 9 and lo), becomes impractical because of the exces-
sive number of experiments required. The method suggested is also more stable
during optimization than the gradient methods which are very sensitive to the
selection of steps in the direction of the gradient and may thus easily shoot out.
of the optimized region. It is also anticipated that other direct search methods,
particularly simplex design, may be easily adapted for optimization of medium
composition. It is obvious that direct search methods may find an application
in process improvement in antibiotic and related industries.
 COMMUNICATIONS TO T H E EDITOR 1837

References
1. D. M. Himmelblau, Process Analysis by Statistical Methods, Wiley, New
York, 1970, chap. 8, p. 230.
2. G. E. P. Box and K. B. Wilson, J. Roy. Stat. Soc., B, 13, 1 (1951).
3. R. Hooke and T. A. Jeeves, J. Assn. C m p . Mach., 8, 212 (1961).
4. H. H. Rosenbrock and C. Storey, Computational Techniques for Chemical
Engineers, Pergamon Press, Oxford, 1970.
5. M. J. D. Powell, Comput. J., 7, 155 (1964).
6. W. Spendley, G. R. Hext, and F. .R. Himsworth, Technometrics, 4, 441
(1962).
7. 0. Volfov4 and P. PilBt, Folia Microbiol. (Prague), 19, 945 (1974).
8. Pil4t and A. Prokop, submitted to Biotechnol. Bioeng., 17, 1717 (1975).
9. J. Auden, J. Gruner, J . Nuesch, and F. Kniisel, Pathol. Microbiol., 30,
858 (1967).
10. K.-D. Schroder and H. Weide, Biotechnol. Bioeng. Symp. No. 4, 713 (1974).
B. Sikyta, A. Prokop, and M. NovAk, Eds., Wiley-Interscience, New York, 1974,
p. 713.

J. VOTRUBA
P. P I L ~ T
A. PROKOP

Dept. of Technical Microbiology

Institute of Microbiology, Czechoslovak
Academy of Sciences
142 20 Prague 4, Czechoslovakia

Accepted for Publication June 30, 1975

View publication stats