- Meeting ID: 867 6409 6440
- passcode: 149120
13h00 - 13h50 – Maxime Mahout (IRD)
Enumerating EFMs of interest with constraint programming methods
ABSTRACT Metabolism is at the heart of the study of biological systems. Metabolic modelling typically involves simulating flux going through reaction networks assuming a steady-state hypothesis. Elementary Flux Modes (EFMs) are the set of subset-minimal metabolic pathways that can be computed from the metabolic network. Such pathways are useful for describing the behaviour of the cell at steady-state, and in particular they can be utilized to uncover reactions essential for cellular functions of interest, or to decompose optimal pathways for cellular growth into elementary bricks. EFMs are expressed as a hybrid logic-linear constrained enumeration problem under the Answer Set Programming (ASP) formalism of aspefm. This constraint programming approach we devised allows us to incorporate genetic regulation, growth medium and thermodynamic constraints and to perform a fast, informative and guided enumeration of EFMs of interest. This enumeration is further helped by network compression and size constraints for the underlying ASP solver clingo[LP]. Furthermore, we explored the performance of constraint programming solvers of different natures across four modelling paradigms: MiniZinc, Prolog, ASP, XCSP in calculating any or all EFMs. We present a benchmark of 50 metabolic networks, 30 of which are integer-valued with traditional tools succeeding in calculating EFMs, and 20 of which are float-valued with only partial enumeration being computationally tractable. The tested tools were able to enumerate integer-valued EFMs of the integer-valued networks, however almost all struggled when confronted to the real-valued networks. We were successful in achieving subset-minimal enumeration of real-valued EFMs with three solvers: clingo[LP] for ASP - aspefm, SWI with modeling library, clpr and clpfd for Prolog and scip for MiniZinc. Out of those, only aspefm and MiniZinc-scip scaled to the largest float-valued metabolic networks.
Dernière modification le 09/01/2026