Section: New Results

Data and knowledge integration based on combinatorial optimization

Participants : Meziane Aite, Lucas Bourneuf, Marie Chevallier, Damien Eveillard, Clémence Frioux, Jeanne Got, Julie Laniau, François Moreews, Jacques Nicolas, Anne Siegel.

A transcriptome multi-tissue analysis identifies biological pathways and genes associated with variations in feed efficiency of growing pigs Our work on the identification of upstream regulators within large-scale knowledge databases (prototype KeyRegulatorFinder) [59] was valuable for figuring out the main gene-regulators of the response of porks to several diets [F. Moreews, A. Siegel] [18]

FCA in a Logical Programming Setting for Visualization-oriented Graph Compression We have explored the underlying idea of lossless network compression to address the problem of uncertainty in biological networks built from predictions, to help to visualize the networks and to classify their nodes in accordance with available annotations [119]. Network compression has been used with success in Dresden (M. Schroeder) with a heuristic approach called Power Graph analysis building abstract graphs where nodes are clusters of nodes in the initial graph and edges represent bicliques between two sets of nodes. First encouraging results have been presented (best paper award) showing that it is possible to mimic the Power Graph behaviour while opening the possibility to achieve better compression levels compared to alternative compression schema. [L. Bourneuf, J. Nicolas] [24]

Metabolic network completion and analysis We released the application paper of the tool Meneco, a tool dedicated to the topological gap-filling of genome-scale draft metabolic networks. The tool reformulates gap-filling as a qualitative combinatorial optimization problem, omitting constraints raised by the stoichiometry, and solves this problem using Answer Set Programming. Run on an artificial test set of 10,800 degraded Escherichia coli networks, we evidenced that Meneco outperforms the stoichiometry-based tool Gapfill in terms of precision. In addition, Meneco reports 10 times less putative reactions than MILP-based tool Fastgapfill for an equivalent precision. This is a strong advantage for manual curation post-processing, since curating 50 to 80 reactions is still possible whereas manually-curating 800 reactions is out-of-range. Meneco was applied to the reconstruction and understanding of a pathogeneic strain of salmon. [C. Frioux, J. Got, A. Siegel] [21], [16]

Toward the study of metabolic functions in communities of organisms In [21], we provided a first example on how to use topological metabolic modeling to assess the complementarity between two members of an algal ecosystem. Since this study, we generalized the selection of subcommunities of interest and propose likely interactions that could occur between seaweeds and their associated bacteria. A focus has also been done on plant microbiota and the reasons underlying the organization of the community. Altogether, these on-going works enable a better understanding of holobiont organizations and functioning. [M. Aite, M. Chevallier, C. Frioux, J. got, A. Siegel, C. Trottier] [21], [31], [30]

Hybrid Metabolic Network Completion In order to improve the precision of gap-filling approaches, we introduced a hybrid approach to formally reconcile existing stoichiometric and topological approaches to network completion in a unified formalism. An hybrid ASP encoding based on MILP constraint propagator was developed. It relies upon the theory reasoning capacities of the ASP system Clingo to solve the resulting logic program with linear constraints over reals. For short, this technology made it possible to combine the best of the combinatorial problem solver Clingo with the MILP solver CPlex. Run on the artificial test set of 10,800 degraded Escherichia coli networks introduced in [21], our approach yielded greatly superior results than obtainable from purely qualitative or MILP approaches. [C. Frioux, A. Siegel] [26], [19]

Combining graph and flux-based structures to decipher phenotypic essential metabolites within metabolic networks Whenever flux or graph-based criteria are used to study metabolic networks, these analyses are generally centered on the outcome of the network and considers all metabolic compounds to be equivalent in this respect. We generalized the concept of essentiality to metabolites and introduced the concept of the phenotypic essential metabolite (PEM) which influences the growth phenotype according to sustainability, producibility or optimal-efficiency criteria. The exhaustive study of phenotypic essential metabolites in six genome-scale metabolic models suggests that the combination and the comparison of graph, stoichiometry and optimal flux-based criteria allow some features of the metabolic network functionality to be deciphered by focusing on a small number of compounds. [C. Frioux, J. Laniau, A. Siegel] [19]