Deciphering regulatory relationships with a logic-based model of causality for gene expression associations

Andrés Aravena[1][2], Carito Guziolowski[3], Max Ostrowski[4], Torsten Schaub[4], Damien Eveillard[2][5], Alejandro Maass[1][6], Anne Siegel[2][7]

Abundant evidence shows that abiotic stresses strongly affects gene expression, yielding different forms of gene synchronization. A natural explanation is the existence of shared transcriptional regulators. However, the discovery or complete characterization thereof remains a challenging problem. State-of-the-art bioinformatics regulation discovery methods are commonly based on the prediction of putative regulator genes and their targets, together with different ways of integrating predictions of co-regulated genes from expression data. Nevertheless, in general, this integration generates lists of results which are orders of magnitude greater than experimentally validated data and needs thorough post processing.

This work proposes a method that integrates predictions of transcription factors, binding sites and operons with gene associations induced by transcriptomic data in order to produce a realistic regulatory graph. This graph results from the solution of a combinatorial problem implemented using Answer Set Programming, a logic-based paradigm which enables an effective encoding and solving of complex combinatorial problems. Our approach, benchmarked on E. coli, provided a regulatory graph that recovers essential features of the gold standard regulatory network for this organism, keeping experimentally validated regulations with significantly higher probability than non-validated ones. In addition, it shares the expected topological properties of a regulatory network. As a major functional output, our approach can be used to highlight functional relationships between genes clustered together in transcriptomic experiments but moreover emphasizes within the whole genome the key functional global regulators which are necessary when the bacterial system is stressed by environmental conditions.

Try Lombarde online on the Mobyle platform at, under the network utilities.

Lombarde supplementary material

(logic based method for regulation deciphering)

The encoding is interpreted using Potassco implementation of ASP solver as follows:

gringo generateExplanations.lp associations.lp input-graph.lp | clasp 0


  1. Center for Mathematical Modeling (UMI-CNRS 2807) and Center for Genome Regulation, University of Chile, Chile  ↩

  2. INRIA, Centre Rennes-Bretagne-Atlantique, Project Dyliss, Campus de Beaulieu, Rennes, France  ↩

  3. École Centrale de Nantes, IRCCyN (UMR CNRS 6597), Nantes, France  ↩

  4. University of Potsdam, Germany  ↩

  5. LINA (UMR 6241), Université de Nantes, Ecole des Mines de Nantes & CNRS, Nantes, France  ↩

  6. Department of Mathematical Engineering, University of Chile, Chile  ↩

  7. CNRS UMR 6074, IRISA Project Dyliss, Université de Rennes 1 (UMR 6074), France  ↩