convert parsimonious wrappers to the other QP ones

src/analysis/parsimonious_flux_balance_analysis.jl

@@ -33,13 +33,21 @@ using the callback in `qp_modifications`, which are applied after the FBA. See
 the documentation of [`flux_balance_analysis`](@ref) for usage examples of
 modifications.
 
-Thhe optimum relaxation sequence can be specified in `relax` parameter, it
+The optimum relaxation sequence can be specified in `relax` parameter, it
 defaults to multiplicative range of `[1.0, 0.999999, ..., 0.99]` of the original
 bound.
 
 Returns an optimized model that contains the pFBA solution (or an unsolved model
 if something went wrong).
 
+# Performance
+
+This implementation attempts to save time by executing all pFBA steps on a
+single instance of the optimization model problem, trading off possible
+flexibility. For slightly less performant but much more flexible use, one can
+construct parsimonious models directly using
+[`with_parsimonious_objective`](@ref).
+
 # Example
 ```
 model = load_model("e_coli_core.json")

src/wrappers/MinimizeDistance.jl

@@ -25,7 +25,21 @@ variants of the same concept. Internally powered by
 [`MinimizeSolutionDistance`](@ref).
 """
 minimize_solution_distance(center::Vector{Float64}) =
-    model -> MinimizeSolutionDistance(center, model)
+    model::AbstractMetabolicModel -> MinimizeSolutionDistance(center, model)
+
+"""
+$(TYPEDSIGNATURES)
+
+Set an objective that finds a solution of minimal norm. Typically, it is
+necessary to also add more constraints to the objective that prevent optimality
+of the trivial zero solution.
+
+This can be used to implement [`parsimonious_flux_balance_analysis`](@ref) in a
+flexible way that fits into larger model systems.
+"""
+with_parsimonious_objective() =
+    model::AbstractMetabolicModel ->
+        MinimizeSolutionDistance(zeros(n_variables(model)), model)
 
 """
 $(TYPEDEF)
@@ -41,13 +55,14 @@ end
 
 Accessors.unwrap_model(m::MinimizeSemanticDistance) = m.inner
 
-Accessors.objective(m::MinimizeSemanticDistance) =
-    let
-        Sem = variable_semantic_mtx(m.semantic, m.inner)
-        Sem' *
-        [spdiagm(fill(-0.5, size(Sem, 2))) m.center] *
-        [Sem zeros(size(Sem, 1)); zeros(size(Sem, 2))' 1.0]
-    end
+function Accessors.objective(m::MinimizeSemanticDistance)
+    (_, _, _, smtx) = Accessors.Internal.sematics(m.semantic)
+    Sem = smtx(m.inner)
+
+    return Sem' *
+           [spdiagm(fill(-0.5, size(Sem, 2))) m.center] *
+           [Sem zeros(size(Sem, 1)); zeros(size(Sem, 2))' 1.0]
+end
 
 """
 $(TYPEDSIGNATURES)
@@ -59,7 +74,20 @@ variant, and [`minimize_solution_distance`](@ref) for working directly upon
 variables. Internally powered by [`MinimizeSemanticDistance`](@ref).
 """
 minimize_semantic_distance(semantic::Symbol, center::Vector{Float64}) =
-    model -> MinimizeSolutionDistance(semantic, center, model)
+    model::AbstractMetabolicModel -> MinimizeSolutionDistance(semantic, center, model)
+
+"""
+$(TYPEDSIGNATURES)
+
+Set an objective that finds a solution of minimal norm in a given semantic.
+This can be used to implement various realistic variants of
+[`parsimonious_flux_balance_analysis`](@ref).
+"""
+with_parsimonious_objective(semantics::Symbol) =
+    model::AbstractMetabolicModel -> let
+        (_, n_sem, _, _) = Accessors.Internal.semantics(semantics)
+        MinimizeSemanticDistance(semantic, zeros(n_sem(model)), model)
+    end
 
 """
 $(TYPEDEF)
@@ -92,4 +120,14 @@ for simpler variants. Internally powered by
 [`MinimizeProjectedDistance`](@ref).
 """
 minimize_projected_distance(proj::SparseMat, center::Vector{Float64}) =
-    model -> MinimizeProjectedDistance(proj, center, model)
+    model::AbstractMetabolicModel -> MinimizeProjectedDistance(proj, center, model)
+
+"""
+$(TYPEDSIGNATURES)
+
+Set a quadratic objective that minimizes the norm in a given projection of
+model variables.
+"""
+with_parsimonious_objective(proj::SparseMat) =
+    model::AbstractMetabolicModel ->
+        MinimizeProjectedDistance(proj, zeros(size(proj, 1)), model)

test/analysis/parsimonious_flux_balance_analysis.jl

@@ -19,12 +19,12 @@
     d2 =
         model |>
         with_changed_bound("biomass1", lower_bound = 10.0) |>
-        with_parsimonious_solution(:reaction) |>
+        with_parsimonious_objective(:reaction) |>
         flux_balance_analysis(Clarabel.Optimizer, modifications = [silence]) |>
         values_dict
 
     @test all(isapprox(d[k], d2[k], atol = QP_TEST_TOLERANCE) for k in keys(d2))
 
-    Q = objective(model |> with_parsimonious_solution(:reaction))
+    Q = objective(model |> with_parsimonious_objective(:reaction))
     @test all(Q[i, i] == -1 for i = 1:7)
 end

test/reconstruction/enzyme_constrained.jl

@@ -92,7 +92,7 @@
         model |>
         with_changed_bound("BIOMASS_Ecoli_core_w_GAM", lower_bound = growth_lb) |>
         with_enzyme_constraints(; total_gene_product_mass_bound) |>
-        with_parsimonious_solution(:enzyme) |>
+        with_parsimonious_objective(:enzyme) |>
         flux_balance_analysis(
             Clarabel.Optimizer;
             modifications = [modify_optimizer_attribute("max_iter", 1000), silence],