Evaluation Functions

All evaluation functions/optimization objectives are defined in mewpy.optimization.evaluation module.

Biomass-Product Coupled Yield (BPCY):

The maximization of the Biomass-Product Coupled Yield is one of the most commonly used objectives in Computational Strain Optimization.

from mewpy.optimization.evaluation import BPCY
fevaluation = BPCY(biomass_reaction_id,product_reaction_id)

By default, MEWPy computes reaction yields using pFBA. This may thought be altered by defining an alternative phenotype simulation method, such as lMOMA.

fevaluation = BPCY(biomass_reaction_id,product_reaction_id,method ='lMOMA')

Also, the BPCY computation may account for a carbon source or substrate consumption:

from mewpy.optimization.evaluation import BPCY
fevaluation = BPCY(<biomass_reaction_id>,<product_reaction_id>,\
                   uptake=<substrate_reaction_id>)

Weighted Yield (WYIELD):

BPCY, on its own, has some limitations. Although the BPCY score of a mutated solution may be high, the flux value of the target reaction may be unstable with the max biomass. To guide the EA to more robust solutions, MEWpy also includes a weight yield objective, that encompasses the target product flux variability, constrained to a minimal growth and introduced metabolic modifications.

from mewpy.optimization.evaluation import WYIELD
fevaluation = WYIELD(<biomass_reaction_id>,<product_reaction_id>)

The trade-off parameter is by default set to 0.3. However it may be altered by adding a new parameter when instantiating the class, for example, alpha=0.5.

fevaluation = WYIELD(<biomass_reaction_id>,<product_reaction_id>,alpha=0.5)

The minimum growth yield may be explicitly defined, min_biomass_value=<some_value>, or as a percentage of the wild type biomass, min_biomass_per=0.1, that is 10%.

BPCY with FVA

MEWpy also includes an objective function that combines BPCY and WYIELD, whose formulation is:

from mewpy.optimization.evaluation import BPCY_FVA
fevaluation = BPCY_FVA(<biomass_reaction_id>,<product_reaction_id>,uptake=<substrate_reaction_id>)

As in BPCY, the substrate is optional and fluxes may be obtained using different phenotype simulation methods.

This objective function is based on a proposal from “OptRAM: In-silico strain design via integrative regulatory-metabolic network modeling” where the additional factor to BPCY favors solutions with a smaller gap between the product minimum an maximum FVA.

Product Yield

The most straight forward objective function is the product yield, where the goal is to maximize the production of a target product. In the MEWpy implementation, this objective function may consider a minimum biomass production explicitly defined or a percentage of the wild type growth. This objective function also may consider distinct phenotype simulation methods. Also, this same objective function may be used for the minimization of a targeted reaction flux by changing the optimization sense, that is, by setting the argument maximize=False.

from mewpy.optimization.evaluation import TargetFlux
fevaluation = TargetFlux(<product_reaction_id>)

Minimum number of modifications

Although problems definition allows for setting a maximum number of modifications, and solutions from the final population may be automatically simplified to remove unnecessary modifications, the minimization of the number of perturbations can be set as an additional optimization objective.

from mewpy.optimization.evaluation import MinCandSize
fevaluation =  MinCandSize()

Modification type

This objective favors modifications with deletions and down regulations. As such, a solution that encompasses more deletions is considered better than one with many up regulations.

from mewpy.optimization.evaluation import ModificationType
fevaluation =  ModificationType()

Molecular weight

Minimizes the sum of molecular weights of the products/substrates of a set of reactions (g/gDW/h).

from mewpy.optimization.evaluation import MolecularWeight
fevaluation =  MolecularWeight([r_id_1,r_id_2,...])

Combining two or more objectives

The previously defined objective functions may be combined into a linear aggregated weighed sum and used in single objective optimization algorithms, such as Genetic Algorithm or Simulated Annealing.

Though the sum of all weights should be equal to 1, this is not imposed as weights may also be used to introduce a normalization for each function. When not provided, the aggregated function assigns a same weight to all functions w=1/n.

from mewpy.optimization.evaluation import BPCY, WYIELD, AggregatedSum
f1 = BPCY(biomass_reaction_id,product_reaction_id,method ='lMOMA')
f2 = WYIELD(<biomass_reaction_id>,<product_reaction_id>)

fevaluation = AggregatedSum([f1,f2],tradeoffs=[0.7,0.3])