Genetic Network Inference

Genetic network inference

People: Peter Antal, Geert Fannes, Kathleen Marchal, Janick Mathys, Yves Moreau

Description:
Molecular biological high-throughput techniques allow simultaneous monitoring of temporal changes of the complete transciptome, proteome or metabolome of a cell elicited by external signals. Such high-dimensional data contain a wealth of biological information if analyzed properly. Genetic network inference tries to reconstitute from such experimental data (Fig. 1A) the underlying regulatory interactions responsible for the observed behavior (Fig. 1B).

Fig. 1A Fig. 1B

Objectives:

Developing a generic methodology for genetic network inference in the Bayesian context. Developing such methodology implies solving complex subproblems, adaptation of existing mathematical techniques to biological concepts and integration of these techniques into the complex methodology. A real life biological problem (i.e. constructing a mathematical description of a biological test system) will be used to develop the methodology. Predictions based on a preliminary mathematical description will be used to design subsequent experiments that are most informative for completion of the mathematical description of the biological process (experiment design).
A second biological goal of this project consists of using the developed mathematical description of the biological system as a tool for fundamental research (prediction of yet undiscovered regulatory interactions) and as a rationale for metabolic engineering of the studied test system.

Biological model system:
Theoretically modeling a complete regulatory network necessitates measuring dynamic variations in the levels of mRNA, proteins, and metabolites. Due to experimental limitations, in a first stage only network models describing transcriptional interactions will be developed (i.e. genetic network models). This implies that the state variables of the dynamic system represent the measured mRNA levels. Posttranslational interactions can be considered as unobserved variables. Known interactions can be used to test the performance of the developed inference methodology. For each of the tested systems it will be assumed that the connections between genes are constant during the course of the same dynamic process (i.e. as long as the same genetic network is active) (time-invariant dynamic system). Two biological model systems will be studied:

Regulatory pathway of the type III secretion system of Salmonella typhimurium TTSS (SpI) (link naar website FAJ), responsible for bacterial invasion of epithelial cells in cooperation with the Centre for Microbial and Plant Genetics.
Regulation of the glycolysis in Saccharomyces cerevisiae. (link naar Thevelein) in cooperation with the Laboratory of Physiology and Biochemistry of microorganisms and Plants.

Development of a model in the Bayesian context:

The Bayes theorem indicates how a prior belief in a hypothesis (M) can be converted to a posterior belief p(M|D) by observation of the experimental data. For instance, a biologist is interested in knowing how a set of genes interact (unraveling of the genetic network structure). Based on his prior knowledge the expert has a certain hypothesis on the structure of the genetic network, to which he assigns a certain prior belief. To update this hypothesis new experiments will be performed. Subsequently the probability is calculated that the experimental data are generated by the hypothetical model (likelihood p(D|M)). Via Bayes theorem prior belief and likelihood are converted to the posterior belief i.e. the probability that the proposed hypothetical model represents the real biological model. The posterior belief therefore represents the belief in the model but actualized by the data.
For biological problems such Bayesian approach can be powerful since most biological hypotheses (prior knowledge) are based on previously validated experimental observations.