Professional Documents
Culture Documents
Received 15 August 2004; received in revised form 4 November 2004; accepted 9 November 2004
Available online 15 December 2004
Abstract
Herein, a back-propagation artificial neural network (BP-ANN), a fit and predictive tool and suitable for bridging the inputs and outputs
of a non-linear problem, is used to model the adsorption of bovine serum albumin (BSA) on porous polyethylene (PE) membrane. Based
on the adsorption data from FTIR-mapping, the parameters of the neural network with a hidden layer are determined by a trial-and-error
method. There is a good agreement between the predicted results by BP-ANN and the experimental data, and the interpolative predictions by
BP-ANN are more precise than those by convectional diffusion equation. Though BP-ANN cannot provide detail information concerning the
mechanism like a conventional diffusion equation (which can permit an evaluation of diffusion coefficient or mass transfer coefficient), it is
a powerful predictive tool as well as a useful compensation to conventional diffusion equation when dealing with the problems of which the
mechanism is not entirely understood or very complex.
© 2004 Elsevier B.V. All rights reserved.
Keywords: Adsorption; Modelling; Membrane; Bovine serum albumin; Artificial neural network
0376-7388/$ – see front matter © 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.memsci.2004.11.007
138 R.-Q. Fu et al. / Journal of Membrane Science 251 (2005) 137–144
the calculated and target outputs, and weights are adjusted tion perpendicular to membrane surface. The breadth of the
by back-propagating the error from the output layer to the line sample is equal to the thickness of original membrane
input layer. The adjustment of the weights can be made sample (200 m) and the thickness of line sample (distance
immediately after the error is detected when inputting each between the knife) is about 50 m. The IR measurements
individual data pair (immediate adjustment), or be made after are conducted with a Nic-Plan microscope FTIR instrument
accumulating the individual errors for all data pairs (deferred (with a Magna-IR System 560 and a Nicolet 0.25 MCT-A
adjustment). Herein, the deferred adjustment is used, and IR detector, Nicolet Instrument Co., Madison, WI). The IR
the adjustment between the (n)th and the (n + 1)th iterative system is equipped with a micro X–Y stage (INA/KLANEW,
is [13]: Tokyo, Japan) for two-dimension mapping of the line sam-
ple. The IR data acquisition and microscope mapping are
1
wji (n + 1) = η δk,j outk,i + αwji (n) (3) controlled by a PC using the software of Omnic and Atlus
K (both from Nicolet), respectively. The resolution of IR is
k
4 cm−1 and 100 IR scans are made to obtain an IR spec-
where wji is the weight adjustment between the neuron j trum at each point of the sample. All the spectra are col-
in the considering layer and the neuron i in the former layer, lected in the range of 1000–2000 cm−1 . Each line mapping
outk,i the calculated output of the neuron i for data pair k, contained about 11–13 points across the whole membrane
δk,j the δ term of the neuron j for data pair k calculated from thickness, and included the measurement of a background IR
the calculated and target outputs, η the learning rate, α the spectrum at a point of KBr sample holder near one edge of the
momentum factor for preventing oscillations and K is the sample.
number of data pairs.
Thus, the unknown parameters η, α and neuron number in 3.3. Determination of adsorption data
the hidden layer should be determined in advance. Herein, the
trial-and-error method and the split-sample validation proce- Herein, the PE membrane has no infrared absorbance
dure are used, that is, the sample data pairs are divided into in the Amide I bands (nearby 1650 cm−1 ) and Amide II
training set and validating set, and the parameters and itera- bands (nearby 1550 cm−1 ), and the adsorption concentration
tive number must be taken to make the error of the validating of BSA in membrane phase can be quantitatively analyzed
set minimize. Herein, the error is expressed as root mean by Amide II area because the absorbance intensity in the
squares (rms) error. area has been reported to be proportional to the amount of
protein adsorbed [7,27] and is believed not to be very sen-
sitive to the conformation of the protein [5]. Our prelimi-
3. Experimental descriptions nary experimental results also show that Amide II area is
linearly regressed with adsorption concentration with a re-
3.1. Materials gression factor about 0.992 [10]. As an example, the FTIR-
mapping at the adsorption time 50 min is shown in Fig.
Bovine serum albumin (Lot 113H0247, water con- 2. The peak around 1462 cm−1 corresponds to –CH2 – of
tent = 1.7%, molecular weight = 69,000, isoelectric point PE membranes, while the peaks 1650 and 1550 cm−1 to
(IEP) = 4.9) is obtained from Sigma Chemical Co. and used the Amide I and II bands of adsorbed protein molecules,
without further treatments. Polyethylene membrane (H 2200) respectively. It is clearly observed that, both the intensity
is provided from Asahi Kasei Co. with a thickness 200 m, of Amide I and Amide II peaks symmetrically decrease
porosity 69%, maximum pore diameter 0.28 m and specific with the distance from the membrane surface outside and
area about 18 m2 /g evaluated by BET adsorption. Before ex- with a minimum value at the central point of the mem-
periments, the membrane is cut into square shape with side brane. Thus, the concentration of the adsorption protein is
length about 50 mm and soaked in ethanol for a few min- not only time-related but also depth-related. However, it is
utes and washed with RO water and then equilibrated them hard to read Amide II peak area from FTIR-mapping. To do
with the buffer solution complying to the experiments. Other this, the mapping is further processed and over-stacked in
chemical reagents are used as received.
one plane by Omnic software provided by Nicolet Instru- 4. Results and discussions
ment Co. Fig. 3 shows the over-stacked spectrum for ad-
sorption sample of 50 min (corresponding to Fig. 2). Then, 4.1. The optimization of neural network parameters
the data (Amide II area) is easy to be read from the over-
stacked spectrum by Omnic software for different membrane Here, the adsorption of BSA on porous polyethylene mem-
depth. brane is modelled by a back-propagation artificial neural net-
Our pre-experiments show that the time needed for satura- work with two inputs (nondimensionalized depth and adsorp-
tion adsorption is about 180 min [10] and great changes occur tion time) and one output (nondimensionalized concentra-
at the initial time period. So, adsorption time of the samples tion). A trial-and-error method is used in the optimization
for FTIR-mapping is not more than 50 min, and total 36 data of neural network topological structure. The data pairs are
pairs (the adsorption time ranges from 10 to 50 min with step divided into training set and validating set, i.e., 12 represen-
10 min) are collected. tative data pairs are randomly selected as validating set and
the remaining 24 data pairs are used as the training set. It
3.4. Pretreatment of adsorption data must be concerned that both validating set and training set
contain some data at all adsorption times. According to some
Since the adsorption of BSA on membrane proceeds from preliminary calculations, the change of momentum factor α
the two sides symmetrically, we only need to consider the has a little influence on the validating error when it is set as
adsorption in membrane from the surface to the middle point. from 0.80 to 1.00, but the best validating and training results
For easy processing, herein the adsorption concentration C are obtained as α is set as 0.90. However, the learning rate η
and membrane depth x (the middle position is set as 0) are has apparent impacts on the validating error, so several tests
nondimensionalized by [11] are carried out. At last η is determined at 0.70 to train neural
network. Once the learning rate η and momentum factor α
C − Cg are determined, the optimal hidden node number and iterative
θ= (4)
C0 − C g number could be easily found. Fig. 4 is the dependencies of
and validating error and the optimal iterative number on hidden
node number. Herein, the optimal iterative number means
x
r= (5) the minimal validating error (i.e., training is stopped when
d the error in the validating set begins to increase). As shown
where θ and r are especially the nondimensionalized con- in Fig. 5 (as an example), the training error persistently drops
centration and depth, Cg the saturated concentration, C0 the during the training, while the validating error firstly drops
initial concentration and d is the half of membrane thickness. then increases with iterative number. Thus, the optimal it-
Note that θ is defined so that it is unity in the beginning and erative number can be determined from the validating error
falls down with the time. Herein, the adsorption isotherm at curve. It can be found from Fig. 4 that the validating error
pH = 5.05 is fitted to a Langmuir-type equation to give the tends to be a flat and there is little change when hidden node
saturated amount about 3.88 mg/m2 , C0 is equal to zero and number is larger than 5, however the optimal iterative num-
d is equal to 100 m. ber increases sharply with hidden node number. Thus, as a
R.-Q. Fu et al. / Journal of Membrane Science 251 (2005) 137–144 141
Fig. 4. The dependencies of validating error and the optimal iterative number
on hidden node number.
Fig. 6. The correlation plot of calculated concentration by BP-ANN vs.
compromise of calculating rate and calculating precision, the experimental concentration. () N = 12, R = 0.9888, rms = 0.0082; ()
N = 24, R = 0.9779, rms = 0.0103.
hidden node number is set as 6 and iterative number is set as
17,280 (the corresponding optimal value). Fig. 6 shows the
correlation plot of calculated concentrations by BP-ANN at at adsorption time of 50 min have some deviations from the
the final parameters versus experimental concentrations. The experimental data (as shown in Fig. 11). As mentioned above,
validating set has the correlation coefficient R = 0.9888 and 50 min is the maximum adsorption time. The above results
rms = 0.0082, and the training set has the correlation coeffi- prove that ANN is good in interpolation but their extrapola-
cient R = 0.9779 and rms = 0.0098. Thus, the prediction by tion capabilities (including prediction in limiting inputs) are
BP-ANN with these parameters meets with the application limited [18]. Thus, it is not surprising that there are some
need. deviations between calculated data and experimental data at
50 min.
4.2. Modelling by BP-ANN and comparing with Moreover, it can be observed that the shape of calculated
convectional diffusion equation curves by BP-ANN varies with the adsorption time. For ex-
ample, at the adsorption time of 20 min, the curve is convex
Figs. 7–10 show the experimental data and the calculated and the tangent decreases with the depth (it should be noticed
data by BP-ANN at the above determined parameters when that the tangent is minus and the absolute value of the tangent
adsorption time is 10, 20, 30 and 40 min, respectively. It can increases with the depth), which implies that the concentra-
be found that there is a good agreement between calculated tion gradient (dC/dr) at membrane surface is much larger
data and experimental data, which proves that BP-ANN is a than that at membrane middle point. However, at the adsorp-
powerful fit and predictive tool. However, the calculated data tion time of 50 min, the curve is nearly linear, which implies
Fig. 5. Dropping curves of validating error and training error during iteration Fig. 7. The calculated data by BP-ANN and the calculated data by convec-
(hidden node number is 6). tional diffusion equation when adsorption time is 10 min.
142 R.-Q. Fu et al. / Journal of Membrane Science 251 (2005) 137–144
Fig. 8. The calculated data by BP-ANN and the calculated data by convec-
tional diffusion equation when adsorption time is 20 min. Fig. 10. The calculated data by BP-ANN and the calculated data by convec-
tional diffusion equation when adsorption time is 40 min.
that there is little difference in the concentration gradient be- and the calculated data by this diffusion equation are also
tween membrane surface and its middle point. Furthermore, shown in Figs. 7–11. It is obvious that BP-ANN is a better
the concentration gradient at membrane surface is larger at tool for interpolative prediction (not prediction in limiting
low adsorption time, however the concentration gradient at inputs). As well known, there are very complex interactions
membrane middle point is larger at high adsorption time. between serum proteins and synthetic membranes [1–5]. The
In order to evaluate the stability of the network, the cal- convectional diffusion equation based on Fick’s second law
culated data by BP-ANN at adsorption time of 20, 25 and cannot fully cover the behaviors of serum proteins in mem-
30 min are shown in Fig. 12. As mentioned above, the time branes. Thus, it is not surprised that there are some deviations
values of experimental data range from 10 to 50 min with step between the calculated data by diffusion equation and experi-
10 min, that is, 25 min is not included in the time values of mental data. However, ANN, a highly interconnected system
training data. However, it can be found from Fig. 12 that the of many processing elements, can handle very complex or
predict curve at 25 min runs between the curve at 20 min and less known interrelationships, and is irrelevant to the mecha-
the curve at 30 min, which proves that the network is stable nism. Thus, ANN is a good tool for bridging the experimental
and predict data is credible. data with the experimental conditions when dealing with the
In our previous paper [11], the experimental data are also adsorption of serum proteins on synthetic membranes.
fitted based on the convectional diffusion equation However, as above mentioned, ANNs have poor extrap-
olation capability, and the predictive results by BP-ANN at
∂θ(r, t) D ∂2 θ(r, t) adsorption time 50 min have more deviations than those by
= 2 (6)
∂t d ∂r 2 convectional diffusion equation. At the same time, ANN can-
Fig. 9. The calculated data by BP-ANN and the calculated data by convec- Fig. 11. The calculated data by BP-ANN and the calculated data by convec-
tional diffusion equation when adsorption time is 30 min. tional diffusion equation when adsorption time is 50 min.
R.-Q. Fu et al. / Journal of Membrane Science 251 (2005) 137–144 143
References
[22] C.S. Yin, W.M. Guo, W. Liu, W. Zhao, Z.X. Pan, Estimation and pre- [25] K.I. Funahashi, On the approximate realization of continuous
diction of gas chromatography retention indices of hydrocarbons in mapping by neural networks, Neural Netw. 2 (1989) 183–
straight-run gasoline by using artificial neural network and structural 192.
coding method, Chem. Res. Chin. Univ. 17 (2001) 31–40. [26] Y. Takahashi, Generalization and approximating capabilities of
[23] D.E. Rumelhart, G.E. Hinton, R.J. Williams, Learning representations multilayer networks, Neural Comp. 5 (1993) 132–139 (Taka-
of by back-propagation errors, Nature 323 (1986) 533–536. hashi).
[24] J. Zupan, J. Gasteiger, Neural networks: a new method for solving [27] W.K. Surewicz, H.H. Mantsch, New insight into protein secondary
chemical problems or just a passing phase?, Anal. Chim. Acta 248 structure from resolution-enhanced infrared spectra, Biochim. Bio-
(1991) 1–30. phys. Acta 952 (1988) 115–130.