On deducing causality in metabolic networks

BACKGROUND: Metabolic networks present a complex interconnected structure, whose understanding is in general a non-trivial task. Several formal approaches have been developed to support the investigation of such networks. One of the relevant problems in this context is the comprehension of causality dependencies amongst the molecules involved in the metabolic process. RESULTS: We apply techniques from formal methods and computational logic to develop an abstract qualitative model of metabolic networks in order to determine possible causal dependencies. Keeping in mind both expressiveness and ease of use, we aimed at providing: i) a minimal notation to represent causality in biochemical interactions, and ii) an automated tool allowing human experts to easily vary conditions of in silico experiments. We exploit a reading of chemical reactions in terms of logical implications: starting from a description of a metabolic network in terms of reaction rules and initial conditions, chains of reactions, causally depending one from the another, can be automatically deduced. Both the components of the initial state and the clauses ruling reactions can be easily varied and a new trial of the experiment started, according to a what-if investigation strategy. Our approach aims at exploiting computational logic as a formal modeling framework, amongst the several available, that is naturally close to human reasoning. It directly leads to executable implementations and may support, in perspective, various reasoning schemata. Indeed, our abstractions are supported by a computational counterpart, based on a Prolog implementation, which allows for a representation language closely correspondent to the adopted chemical abstract notation. The proposed approach has been validated by results regarding gene knock-out and essentiality for a model of the metabolic network of Escherichia coli K12, which show a relevant coherence with available wet-lab experimental data. CONCLUSIONS: Starting from the presented work, our goal is to provide an effective analysis toolkit, supported by an efficient full-fledged computational counterpart, with the aim of fruitfully driving in vitro experiments by effectively pruning non promising directions.