Professional Documents
Culture Documents
9 de septiembre de 2018
Autor: Laura Johana Castellanos Suárez
Análisis Crítico - Artículo 2
Modeling bioprocess scale-up utilizing regularized linear and logistic regression
Muhammad Farhan, Antti Larjo, Olli Yli-Harja, Tommi Aho
IEEE INTERNATIONAL WORKSHOP ON MACHINE LEARNING FOR SIGNAL PROCESSING –
2013
The scale-up methodology was consisted of four steps: encoding of categorical variables
to incorporate them into modeling, product yield prediction using regularized linear
regression, yield correspondence-based data rearrangement and scale-up modeling using
regularized linear and logistic regression. Experiments were performed in flasks (81
samples), 2L fermenters (24 samples), and 30L fermenters (12 samples). Each sample
was composed of around 40 typical bioprocess variables such as strain, broth medium,
broth adsorbent, fermentation time, temperature, agitation, aeration, pH, etc. After
dummy coding the categorical variables, the number of variables increased to more than
70. All the samples were exploited when developing the yield prediction model using
regularized linear regression. Then 10-fold cross-validation was performed to determine
the optimal model coefficients based on the minimum prediction error so that the model
is general enough to predict the product yield for an unseen sample.
The conclusion was that the scale-up was successfully achieved from flask to 30L
fermenter, as well as from 2L to 30L fermenter. Instead of these lab-size fermentations,
Análisis Crítico - Artículo 2 | 9 de septiembre de 2018
It was not clear if the authors used their own software or code, or commercial software to
develop the models, optimize or even analyze the results. In the same way, I think that in
order to give more credibility to this new methodology, they should have developed a
greater number of experiments in the 30 L bioreactor because the experiments were less
than in the small ones.
In think the paper content had deep optimization and statistics themes. Therefore, it was
necessary to make a careful reading of it. Likewise, in my opinion if you want to apply this
methodology you should study a lot about the methods proposed to get successful
results. On the other hand, the results were surprising as a consequence of the detailed
mathematical development. I think it could be a breakthrough in the bioprocesses scale-
up and in saving costs and time for companies.