Fig. 3: Simulation study using CORNETO’s capabilities for multi-sample FBA modelling.
From: Unifying multi-sample network inference from prior knowledge and omics data with CORNETO
a, The number of different reactions selected using the single-sample sFBA (orange) versus CORNETO’s multi-sample sFBA (blue), constraining the optimal production of biomass to be a minimum (min.) of 90% (top), 50% (middle) and 10% (bottom) of the optimal attainable biomass for each sample, and for four different scenarios considering 2, 4, 8 and 16 samples. Each sample represents a single, simulated knockout (condition). For each scenario, we generated ten independent subsets of the specified size by randomly sampling conditions from the full pool of knockouts (n = 10 per boxplot). b, The average selection proportion of each reaction across different sample sizes (2, 4, 8 and 16 samples), where 1 indicates selection of the reaction in every case. Results are shown for the top 30 reactions with the greatest differences in selection frequency between the multi-sFBA and single-sFBA methods, under the minimum 90% biomass constraint setting. c, The performance of the sparse flux adjustment method leaving out different percentages of metabolic fluxes across samples, and under different Gaussian noise, for different regularization strengths (λ). The vertical dashed line indicates the baseline results for λ = 0.
