Biology Reference
In-Depth Information
Design of Experiments
Circuit:
l 1
l 2
l 3
Design
Output
l 1
Tune
inputs
Legend
Input 2
l 2
Constrained
library of parts
l 3
100%
Screening and optimization
3
Output
β 0
β i I i
l 2
Part activity
i
0
l 1
0%
Genetic
variant
Optimized
pathway
Input 2
FIGURE 4.4
Design of experiments can be performed if libraries of parts with predictable gene expression are available. An illustrative circuit composed by three genes
is depicted in the figure. Design of experiments provides a formal and efficient way to browse solution space. However, the ability to do that requires that
input parameters can be reliably manipulated. Constrained libraries of parts that provide reliable function can be used to efficiently search the space of
solutions. Design of experiments also enables continuous loop between design and analysis phases, wherein the next round of experiments is driven by the
analysis of previous experiments.
described above limit the number of variants that need to be searched by targeting variation
only to the key parts that most likely influence the desired function, statistical design of
experiments (DoE) can reduce the space of search even further.
75
DoE is a general framework that integrates the planning and analysis phases. It describes a
systematic approach to designing a curated minimal set of experiments that produce the
maximum amount of relevant information about the impact of multiple factors in the
response of a system. DoE is particularly useful in screening and optimization, where the
former identifies the most influential factors and their ranges, and the latter defines the
combination of factors that give optimal response. Initial planning of experiments involves
three steps: first, define the set of factors that can impact the response variable; second,
define the range of levels for each factor; third, define a model, usually polynomial with
interaction between factors, to relate factors with response variable.
The use of DoE is widespread across a variety of applications that include pharmaceutical
research, 52 chromatography optimization, 53 and recombinant protein production. 54 There
are proposals to use it for biological model discrimination and parameterization as well. 55
Though DoE has not been routinely applied to genetic circuit design, it has much potential
for this purpose. Indeed, metabolic pathway circuit optimization is perfectly suited for DoE.
Take the hypothetical process described in Figure 4.4 . As shown, one can use a screening
approach to discover factors (or inputs) that have the greatest influence on a response
(e.g . the concentration of a desired end-product) in a metabolic pathway composed of three
enzymes (inputs). Here, the availability of well-characterized parts libraries that enable
precise manipulation of the circuit parameters (i.e . concentration of enzymes) is crucial.
DoE can then provide a minimal set of experiments that is feasible to test in the laboratory
and will most effectively populate the large space of solutions. There are a number of
different ways to choose the initial set of trials, each informative in a different way.
For example, in a system composed of three factors each with five putative levels of enzyme
concentration, we could use DoE and the central composite design to test experimentally an
Search WWH ::




Custom Search