A journal version is available at https://hal.inria.fr/hal-01153704

Synthetic engineering of bacteria aims at genetically modifying existing bacteria strains in order to turn them into efficient factories for target metabolites, such as biofuels or medicals. Thereby, computational models are used to predict the effects of modifications and by this to enable a more directed search for promising modification strategies.

In this work, we present a new computational modeling approach to predict genetic modification strategies for the overproduction of antimicrobial non-ribosomal peptides (NRPS), e.g. surfactin and mycosubtilin, in Bacillus subtilis [5]. In particular, we are interested in changing the metabolism of B. subtilis in a way, such that the production of NRPS precursors (amino acids) is increased, as there the major bottleneck of NRPS assembly is expected. Our approach bases on models in terms of metabolic and regulation graphs as presented by [3] that are known to provide extended information on regulatory control. We formalize the given graphs using the common kinetic reaction model in terms of ordinary differential equations and apply the assumption of homeostasis, where reactions are considered to be constant fluxes, i.e. derivations are set to 0. In order to deal with the usual problem of big solution spaces as it is resulting from lacking kinetic information, we develop new methods following the paradigm of abstract interpretation [1]. These transform kinetic models into sets of constraints over abstract domains. Constraints are then solved with novel, tailor-made algorithms that base on the ideas of constraint programming [7]. We further assign quality weights to solutions, such that solution sets, which can still be large, are ranked w.r.t. the effectiveness of the underlying modification strategy. This is efficiently integrated into constraint solving by a branch-and-bound approach [2]. Our prediction approach is described in more details in [9].

In comparison to existing methods based on linear optimization [4,8], we circumvent the use of the usual biological assumptions to reduce solution spaces that are considered to be often inappropriate [6]. On the other hand, whereas linear optimization methods rather appear to be black boxes, constraint solvers inherently deliver explanation for the derivation of solutions. First predictions for the overproduction of surfactin based on our approach are currently tested in wet-lab experiments with the B. subtilis 168 derivated strains.

References

[1] P. Cousot and R. Cousot. Basic Concepts of Abstract Interpretation. Kluwer Academic Publishers, 359-366, 2004.

[2] R. J. Dakin. A tree-search algorithm for mixed integer programming problems. The Computer Journal, 8(3):250-255, 1965.

[3] A. Goelzer, et al. Reconstruction and analysis of the genetic and metabolic regulatory networks of the central metabolism of Bacillus subtilis. BMC Systems Biology, 2(1):20+, 2008.

[4] J. Kim and J. Reed. OptORF: Optimal metabolic and regulatory perturbations for metabolic engineering of microbial strains. BMC Systems Biology, 4(1):53+, 2010.

[5] V. Leclère et al.. Mycosubtilin Overproduction by Bacillus subtilis BBG100 Enhances the Organism's Antagonistic and Biocontrol Activities. Applied and Environmental Microbiology, 71(8):4577-4584, 2005.

[6] S. Ranganathan,et al.. OptForce: An Optimization Procedure for Identifying All Genetic Manipulations Leading to Targeted Overproductions. PLoS Comput Biol, 6(4):e1000744+, 2010.

[7] F. Rossi, et al.. Handbook of Constraint Programming. Elsevier, 2006.

[8] N. Tepper and T. Shlomi. Predicting metabolic engineering knockout strategies for chemical production: accounting for competing pathways. Bioinformatics, 26(4) :536-543, 2010.

[9] Mathias John, et al. Knockout Prediction for Reaction Networks with Partial Kinetic Information. 2012. http://hal.inria.fr/hal-00692499 .