MCdesign

class MCdesign(prodnet, number_cores)

This class interfaces the metabolic model simulations of prodnet with the multiobjective optimization methods.

MCdesign(prodnet, number_cores)

Parameters:prodnet (Prodnet class) –

we are interested in solving.

number_cores (Integer, optional). Number of processes to be spawned

by parpool. Defaults to feture(‘numCores’).

N_OBJ_DIGITS = None

(integer) Number of significant digits in objective func, default 6.

OBJ_TOL = None

(double) Corresponds to 1e-obj.N_OBJ_DIGITS

design_parameters = None

(structure) See the method set_design_parameters for details

design_table = None

(Hash table (containers.Map()). Stores previously evaluated individuals. As of Matlab2017, the containers.Map() object data structure is the fastest and most memory efficient option for this task.

enum_parameters = None

(structure) See the method enumerate_alternative_solutions for details

ga_parameters = None

(structure) See the method set_default_ga_parameters for details

max_design_table_size = None

(integer) Limits the number of elements in the design_table to avoid memory issues.

number_cores = None

(integer) Number of processes to be used for parallel worker pool.

prodnet = None

(Prodnet class)

random_number_seed = None

(double) Use a fixed value for reproducible results. This feature has not been throughly tested and it will not work when solving the problem in parallel, since Matlab does not offer yet a universal way to obtain reproducible random numbers in parfor loops.

start_point = None

(structure) See the method set_start_point for details

use_module_variable = None

(logical) Indicates if module reactions are used or not.

vector_difference_tolerance = None

(double) Numerical tolerance for vectors.

add_to_design_table(obj, variable_vector, objective_vector)

Records a solution in the design hash table

Parameters:

• variable_vector (logical vector) – key.
• objective_vector (vector) – value.

calc_enum_penalty_obj_fun(obj, x, excluded_solutions)

Computes penalty objective function. Adapted from method calc_penalty_obj_fun.m

Parameters:

• x (logical vector) – design variable vector.
• excluded_solutions (logical matrix) – Every row corresponds to a x which is not allowed in the final solution.

calc_penalty_obj_fun(obj, x)

compute penalty function objective values

Parameters:x (logical vector) – design variable vector.

combine_module_variables(y, Z)

Combines deletion variables and module variables for all networks into one vector

Parameters:

• x (logical vector) – design variable vector.
• Z (logical matrix) – rows correspond to production networks, and columns to module variables.

create_initial_population(obj, nvars)

Create an initial population for the MOEA

Parameters:nvars (int) – Number of variables.

create_knowledge_based_initial_population(obj, max_deletions, deletion_id, varargin)

Generates a start point for the optimization algorithm based on prior information about useful deletions.

deletion_id (cell): Ids of reactions used to build initial point. population_size(double, optional): Default is obj.ga_parameters.population_size seed(double, optional): Random number generator seed. Default is 1000,

so unless changed the output is deterministic.

Notes

• Currently sampling works by first enumerating all solutions, so the

  the number of candidate reactions and the number of deletions shoudl not be to high.

• Module variables are ignored and will be filled up with zeros by

  src.@MCdesign.create_initial_population().

• Once all combinations of deletions up to the target cardinality have

  been included, the remaining population will be filled with random designs to hit the target size.

crossover_module_variable(obj, parents, ~, nvars, ~, ~, thisPopulation)

Customized genetic operator for gamultiobj().

Parameters:

• parents – Row vector of parents chosen by the selection function
• options – options structure
• nvars – Number of variables
• FitnessFcn – Fitness function
• unused – Placeholder not used
• thisPopulation – Matrix representing the current population. The number of rows of the matrix is Population size and the number of columns is Number of variables.

Returns:

the crossover offspring as a matrix where rows correspond to the children. The number of columns of the matrix is Number of variables.

Return type:

xoverKids

Notes

It seems like this function is slower than the default ( I have not done rigorous timing though).

To optimize potentially one could avoid making copies of variables and calling external functions, but due to the complexity of the design variable vector that could look very ugly.

enumerate_alternative_solutions(obj, enum_parameters, total_max_time)

Enumerates alternative solutions and saves resulting file to problem directory (proble-name/output/alternative_solutions).

Parameters:

• enum_parameters.solution_id (str) – id of the mop_solution file containing the target design to be enumerated.
• enum_parameters.max.deletions (int) – Maximum number of deletions allowed for feasible solutions
• enum_parameters.max.module (vector of int) – Length corresponds to production networks. Maximum number of module reactions allowed in feasible solutions.
• total_max_time (double, optional) – overall time given to the method, default is intmax. Note that the maximum time of each run can be controlled by the parameter obj.ga_parameters.max_time;

Notes

• The genetic algorithm parameters are taken from obj.ga_parameters.

• Regarding enumeration of alternative solutions: 1) for efficiency

  can stop as soon as violation occurs, instead of computing all objectives. 2) it may be intersting to prespecify starting excluded solutions.

enumerate_alternative_solutions_one_target(obj, excluded_solutions)

looks for one alternative solution until it is found or the maximum number of generations is reached. Adapted from solve_mop_for_stall_gen for single objective problem.

Parameters:excluded_solutions (logical matrix) – Rows correspond to designs and columns to design variables.

extract_alternative_solutions(obj, mop_solution)

Finds alternative solutions present in the mop_solution, extracts them, and introduced them into the structure alternative_solutions, which is then appended to mop_solutions.

extract_module_variables(x, n_deletion_var, n_prod)

Decomposes a vector of design variable x into a deletion vector y, and a module reaciton matrix Z.

Parameters:

• n_deletion_var (int) – number of candidate reactions.
• x (logical vector) – vector containing all variabels passed to the optimization algorithnm

Returns:

• y (logical vector) – Reaction deletions (length of cand reactions)
• Z (logical matrix) – Matrix of n_prod by cand reactions.

Notes

This function was recycled so that is why I originally introduced

it as a method, but obviously n_deletion_var and n_prod are static properties of the Prodnet class.

find_dominated_rows(A, B, uniqueFlag, verbose, n_significant_digits)

Provides information regarding domination of the rows from matrix A compared to the rows of matrix B. A and B dont need to have the same number of rows.

Parameters:

• A (matrix) –
• B (matrix, optional) – Default is B = A.
• uniqueFlag (logical, optional) – if true (default), only check unique rows in A, and unique rows in B.
• verbose – 2 display lots of information. 1 display some information. 0 display no information (default).

Returns:

• A_dominated_ind (vector) – Indices of rows dominated in A.
• B_dominated_ind (vector) – Indices of rows dominated in B.
• isEqual (logical) – Indicates if A and B are equal.
• dom_relation_A (containers.Map()) – Key is a solution in A and values are solutions in B dominated by that solution in A.
• dom_relation_B (containers.Map()) – Key is a solution in B and values are solutions in A dominated by that solution in B.

growth_to_nongrowth_designs(obj, mop_solution)

Since the candidate set for growth objectives (wGCP, sGCP) is different than that of nongrowth objectives(NGP), in order to use a growth design as a starting point for NGP objective, they must be converted to the appropriate indices.

lookup_design_table(obj, variable_vector)

Obtain a solution from the design table or an indicator that it is not present.

Parameters:variable_vector (logical vector) – key. Returns:

• isPresent (logical) – True if solution is present.
• objective_vector (vector) – value.

mutationuniform_module(obj, parents, ~, GenomeLength, ~, ~, ~, thisPopulation, mutationRate)

Adapted from Matlab’s mutationuniform.m to satisfy module variable constraints. Input and outputs are defined in Matlab’s documentation.

save_mop_solution(obj, mop_solution)

Saves the mop_solution structure into the appropiate output directory.

set_default_ga_parameters(obj)

Starts up the ga_parameters structure with default values.

set_design_parameters(obj, design_parameters)

Setter for basic design parameters.

Parameters:

• design_parameters.obj (string) – design objective, options are ‘wGCP’, ‘sGCP’, and ‘NGP’
• design_parameters.max_deletions (integer) – Maximum number of deletions allowed.
• design_parameters.max_module (vector) – Length corresponds to the number of production networks. Maximum module variables allowed in each production network.

set_start_point(obj, solution_id, problem_path, mop_solution)

Sets a starting point for the genetic algorithm.

Parameters:

• solution_id (string) – id of the solution to be used as starting point.
• problem_path (string, optional) – Problem from which the solution will loaded. Default is obj.prodnet.problem_path.
• mop_solution (mop_solution structure, optional) – Will ignore other input parameters and set this mop_solution as the starting point.

solve_mop(obj, design_parameters, start_point_info, total_max_generations, total_max_time)

Solves multiobjective optimization problem.

Parameters:

• design_parameters (structure) – Input to set_design_parameters()
• start_point_info (structure, optional) – Input to set_start_point()
• total_max_generations (integer, optional) – Maximum number of generations allowed regardless of convergence criteria.
• total_max_time (double, optional) – Maximum running time in minutes.

solve_mop_for_stall_gen(obj)

Runs MOEA for a fixed number of generations.