Professional Documents
Culture Documents
DOE-Exercise Capto S
DOE-Exercise Capto S
DOE-Exercise Capto S
Background
Capto S is a chromatographic medium, designed for the capture and intermediate
purification of recombinant proteins and as a second step in the purification of monoclonal
antibodies (MAb). The medium combines high rigidity with high dynamic binding capacity
and fast mass transfer to facilitate faster purification.
Objective
The objective of this investigation is to find the best loading conditions to achieve the highest
dynamic binding capacity at 10% breakthrough (QB10%) when using Capto S in the second
step of MAb purification. The factors investigated are pH, conductivity and sample residence
time.
Data
Task 1
Create a 23 full factorial design with three center-points and enter the response data into the
worksheet.
Task 2
Use the analysis wizard to work through the steps in the data analysis. Start with the default
interaction model to get an overview of the possibilities. What can you say about the model
quality? What measures are available to try to improve the model?
Task 3
Check if the model should be quadratic rather than linear.
Hint: The Square test in the Analysis Wizard is enabled; use this tool and select a square term,
for instance the square term of pH or the square term of Conductivity.
Why should we be careful when we test for quadratic terms in a design that only supports
linear and interactions terms?
Also remove any existing non-significant model term. Does the model improve?
Task 4
In order to resolve any confounding between quadratic terms, we need to augment our
original design so it supports quadratic terms. Go to: File: New: Complement Design.
Do not add any extra center-points in the final step. (Adding additional center-points is the
recommended option, but was not done in this case by the original investigators.)
Use the analysis wizard to create a model for the data of the updated worksheet. Go through
the same workflow as in Task 3. What can you say about the relevance of the three quadratic
terms?
Task 5
Which combination for the factors result in the highest dynamic binding capacity?
Create a 4D contour plot where the first axis is pH, the second axis is Conductivity and the
third is Residence time.
Interpret the contour plots.
Task 6
Use the Optimizer to search for a suitable setpoint at which the response specification is
fulfilled.
Is it possible to define a robust setpoint?
Is it possible to define a design space?
Comment on the result.
Task 2
The Analysis wizard was used to analyze the data. The six basic plots relating to the initial
interaction model are seen below.
The replicate plot shows that the response varies between 0 and 149. The replicate error of
the center-points is relatively small. The response values of the center-points give us the first
indication that the model we are seeking is non-linear and therefore cannot be properly
modeled using a screening design (which only supports linear and interaction terms).
According to the histogram, the response is approximately normally distributed and
therefore do not need to be transformed. The summary of fit plot shows that the model does
not fit the data well. The Q2 value is negative and the model validity is just below 0.25. The
weakness of the model is manifested in the wide confidence intervals of the regression
Task 3
The Square test is a tool that is designed to help the user to detect possible non-linearities.
As seen from the screenshot below, highly significant square terms have been detected. The
problem is that the three possible square terms have been estimated to have exactly the
same numerical value, i.e., the three square terms are confounded. In this case confounding
among the square terms arises because the underlying factorial design does not have
enough experiments to allow the quadratic terms to be fully resolved (a k a ‘unconfounded’)
from one another. So whichever square term is included in the model, identical modeling
statistics will be the result. Experiment with the different square terms to verify that this is the
case!
In the next step, the model with the square term of conductivity was further pruned by
dropping two small interaction terms (i.e., ph*Res and Res*Cond). The resulting model has
excellent statistics and the normal probability plot of residuals looks exemplary.
Make sure the model is the full quadratic model in all three factors expanded by the block
factor ($Bl) by selecting pH, Residence Time, and Conductivity and clicking Squares =>.
The addition of quadratic terms results in the model fitting the data much better. This is a
strong indication that we have quadratic effects in the investigated system. The
complementary experiments are added in the right-hand part of the replicate plot and all
show response values between 50 and about 140. Some skewness of the data can be seen in
the histogram, but we expect the quadratic terms to compensate for any non-linearity.
After complementing the design there is a striking improvement of the model quality with an
R2 of 0.98 and Q2 of 0.81. The coefficient plot reveals that pH and Conductivity have a
quadratic effect on QB10%. We can also see that the Block factor is not significant, indicating
that there is no difference between the two designed blocks (screening and complement).
This means the Block factor can be removed from the model.
Task 5
The largest experimental area that produces the highest dynamic binding capacity is found
when the residence time is set to 6 minutes. Here the pH has to be in the range 4.8 to 5.5 and
the conductivity must be between 5.0 and 11.0 ms/cm. In the next task we will examine how
sensitive this region is to small disturbances in the factor values and other sources of
uncertainty.
In the next step, a search for a robust setpoint was initiated. As seen in the graph field, the
profile of the factor settings of the robust setpoint is less extreme than what´s the case for the
first optimal setpoint. This introduces some flexibility not seen for the optimal setpoint.
The Design Space explorer tool was used to visualize the extent of the design space and the
location of the robust setpoint. The insight is that with the settings:
response specification of QB10% > 120;
uncertainty interval = confidence; and
Conclusions
This exercise shows how a screening design can be executed, evaluated and complemented
to produce an optimization (RSM) design. After the first screening the model was too simple
to describe the behavior of Capto S under different conditions. A complementary
optimization showed that both pH and conductivity had a quadratic influence on the QB10%,
meaning that they both peaked within their specified interval. The response contour plots
showed that a long residence time in combination with high pH and low conductivity
resulted in the highest QB10%. Using the Optimizer, it was possible to define a robust
setpoint corresponding to the settings pH ≈ 5.04, residence time ≈ 5.1 minutes and
conductivity ≈ 7.2 ms/cm.