Extreme Scale-Down Approaches for Rapid Chromatography Column Design and Scale-Up During Bioprocess Development - Measurement, Monitoring, Modelling, and Control of Bioprocesses - page 132

Table 1 Situations in which to use the simplex method and where to employ an alternative

Issue Use the simplex algorithm Use an alternative

Stage of
development

Deploy during early process
development when the main aim
is to head in the right direction
within a design space

Use a factorial design later in
process development, either as
part of a design space
submission to a regulator, or as
the basis of a model for large-
scale process control

Experimental cost
versus value of
process
understanding

Use the simplex method when it is
necessary to extract process
knowledge from very small
quantities of material, thus
maximising the value of every
experiment early in
development. Alternatively, if a
good scale-down model does not
exist and hence when one is
relying upon time- and resource-
consuming operations e.g. at a
large laboratory scale, then the
simplex algorithm allows one to
make best use of that
experimental effort

If feed material availability is not an
issue, or when achieving a good
result from every data point is
not as critical e.g. when one can
afford the occasional misleading
result or experimental failure,
the simplex approach may not
be as relevant; For example, if
highly parallelised microscale
operation on robotic platforms is
being considered, typically this
will generate large quantities of
data and the general trends
within the resulting response
surfaces may be sufficient to
overcome a small number of
failures

Analytical burden When the analytical burden is

significant or time-consuming, it
may be necessary to reduce the
total number of test conditions
to avoid bottlenecks. Thus again
it becomes necessary to
maximise the value of every
data point in an experiment and
the simplex technique can
facilitate this aim

If analysis is fast enough that one
does not need to resort to using
only the bare minimum number
of runs to prevent throughput
bottlenecks, then the simplex
technique may not be as
pertinent

the utility of integrating user expertise into the optimization. A semi-closed loop
enables data evaluation and optimisation during a GA iteration to be regulated by
an operator to allow process understanding to be injected where appropriate. Thus,
the initial experiments defined for the first GA iteration were chosen in part by the
user to ensure good coverage of the entire variable range and to ensure that a large
number of gradient shapes were evaluated during the study in the search for the
optimal multilinear gradient shape.