Source-linked AI summary
Fast Reconstruction of Compact Context-Specific Metabolic Network Models
Nikos Vlassis, Maria Pires Pacheco, Thomas Sauter
TL;DR
Context-specific metabolic models require efficient reconstruction from global networks and high-throughput data. FASTCORE uses strongly supported core reactions and sparse-mode-based linear programs to build flux-consistent subnetworks with minimal additions. In liver experiments, it was several orders of magnitude faster than MBA and produced more compact reconstructions.
Problem
Reconstructing minimal flux-consistent context-specific networks from global models and active core reactions is NP-complete, while practical applications require repeated testing and parameter tuning.
Method
FASTCORE iteratively computes sparse modes through linear programs to find a flux-consistent subnetwork containing all core reactions and a minimal set of additional reactions.
Results
FASTCORE was several orders of magnitude faster than MBA and produced more compact liver reconstructions, differing by 70-80 non-core reactions.
Takeaways & Limitations
FASTCORE's simplicity, lack of free parameters, and speed support repeated runs for testing core sets or optimizing reconstruction parameters.
Takeaways & Limitations
The exact formulation does not scale well because it requires all elementary modes and searches over a binary hypercube of dimension n.
Abstract
from arXiv · showhide
Systemic approaches to the study of a biological cell or tissue rely increasingly on the use of context-specific metabolic network models. The reconstruction of such a model from high-throughput data can routinely involve large numbers of tests under different conditions and extensive parameter tuning, which calls for fast algorithms. We present FASTCORE, a generic algorithm for reconstructing context-specific metabolic network models from global genome-wide metabolic network models such as Recon X. FASTCORE takes as input a core set of reactions that are known to be active in the context of interest (e.g., cell or tissue), and it searches for a flux consistent subnetwork of the global network that contains all reactions from the core set and a minimal set of additional reactions. Our key observation is that a minimal consistent reconstruction can be defined via a set of sparse modes of the global network, and FASTCORE iteratively computes such a set via a series of linear programs. Experiments on liver data demonstrate speedups of several orders of magnitude, and significantly more compact reconstructions, over a chief rival method. Given its simplicity and its excellent performance, FASTCORE can form the backbone of many future metabolic network reconstruction algorithms.
1 Introduction
Context-specific metabolic models select reactions active in a particular cell or tissue, but reconstructing them from global networks and high-throughput data requires mathematically constrained, efficient algorithms. FASTCORE addresses this need by finding flux-consistent subnetworks containing strongly supported core reactions while adding few reactions.
- Genome-scale models combine omics and literature data but contain reactions that may be active in different cellular contexts.
- Context-specific models retain reactions active in a given cell or tissue and can have greater explanatory and predictive power than global models.
- Most reconstruction algorithms first select reactions using omics evidence, then identify a mathematically valid subnetwork containing those reactions.
- FASTCORE starts from a strongly supported core set and searches for a flux-consistent global-network subnetwork containing that core plus a minimal set of additional reactions.Flux consistency requires each reaction to carry nonzero flux in at least one feasible flux distribution.
- FASTCORE uses sparse modes and linear programs to reconstruct models, achieving several-orders-of-magnitude speedups and more compact liver reconstructions than MBA.
2 Methods
The methods define metabolic-network consistency and formulate context-specific reconstruction as finding a smallest consistent subnetwork containing an active core set. FASTCORE approximates this NP-hard search by greedily adding sparse modes computed with linear programs.
- Network representation: A metabolic network is represented by a stoichiometric matrix and feasible flux vectors satisfying steady-state mass-balance constraints and flux bounds.The constraints are expressed as Sv = 0 with v ∈ B.
- Network consistency: A reaction is blocked when it cannot carry nonzero flux in any feasible mode, and a network is consistent when it contains no blocked reactions.Blocked reactions can be identified by testing whether any feasible mode activates them.
- Network consistency: The single-LP consistency test can produce spurious solutions because reversible reactions may be represented by equally fluxed opposing irreversible copies.In the toy network, the test reports z*=1.5 even though A↔B is blocked.
- Context-specific reconstruction: The context-specific reconstruction problem seeks the smallest reaction set containing the core reactions whose induced subnetwork is consistent, but this problem is NP-complete.An exact MILP formulation is possible, but it requires all elementary modes and binary variables indexed by the global reaction set, limiting scalability.
- Context-specific reconstruction: A union of supports from global-network modes induces a consistent subnetwork, allowing reconstruction to be formulated as a search over mode sets.The exact formulation searches for a minimum-cardinality union containing the core set.
- FASTCORE: FASTCORE uses a greedy strategy that iteratively computes modes with sparse support outside the core set, using L1-norm minimization to limit added reactions.Each iteration adds a mode while approximately minimizing the cardinality of the reconstructed network.
- FASTCORE: Unlike MBA, FASTCORE expands the active set from an empty set, whereas MBA starts with the full network and repeatedly prunes reactions.MBA's pruning order can affect its output model, motivating repeated runs.
3 Results
The experiments evaluate fastcc and fastcore for consistency testing and context-specific reconstruction, comparing speed, compactness, approximation quality, and biological relevance across models and validation settings.
- 3.1 Consistency testing: fastcc was faster and used much fewer LPs than fastFVA and CMC across four input models.The comparison included c-Yeast, c-Ecoli, c-Recon1, and c-Recon2.
- 3.2 Reconstruction of a liver model: fastcore reconstructed a liver-specific model in about one second, several orders of magnitude faster than MBA and with fewer LPs.MBA’s reported LP and runtime counts referred to one pruning iteration, while its reconstruction sizes reflected 1000 iterations.
- 3.2 Reconstruction of a liver model: fastcore reconstructions were 70-80 non-core reactions more compact than MBA reconstructions.The standard and strict liver models shared 96% and 95% of fastcore’s reactions with MBA, respectively.
- 3.2 Reconstruction of a liver model: On a small E. coli model, fastcore produced near-optimal reconstructions whose quality improved as the core-set size increased.The comparison used mean reconstructed-model sizes over 50 repetitions for each core set against an exact MILP solver.
- 3.2 Reconstruction of a liver model: Random sub-sampling validation yielded a median p-value of 0.0025, indicating recovery of liver-specific reactions left out of the input core set.The procedure repeatedly reconstructed models from randomly selected subcore sets containing 80% of the original core reactions.
- 3.2 Reconstruction of a liver model: The extended liver model evaluated urea/glutamine ratios under healthy, partial-knockout, and full-knockout conditions.The modeled mutations involved ASS, ASL, and OTC, with reaction rates set to 500, 250, or zero.
- 3.2 Reconstruction of a liver model: A murine macrophage model was built from Recon1bio using 300 proteomics-derived Raw264.7 macrophage reactions as the core set.Recon1bio contains 3745 reactions, while its consistent part used for reconstruction contains 2474 reactions.
4 Discussion
FASTCORE was motivated by the need for fast computation and compact context-specific metabolic reconstructions. Its speed supports repeated analyses, while core-set design can preserve biologically supported alternative pathways despite compactness.
- FASTCORE targets fast computation and compact output models in context-specific metabolic network reconstruction.
- Faster reconstruction permits multiple runs to optimize parameters or test different core sets extracted from input data.This is useful when evaluating how output models change across gene-expression thresholds.
- Compactness need not eliminate biological redundancy when redundant reactions supported by evidence are included in the core set.Alternative pathways can therefore remain in the reconstructed model when biologically justified.