Gene Networks

A popular method to represent gene regulation is to draw network diagrams where genes connect to other genes as if they directly affect each other. Such gene networks are phenomenological models because they do not represent explicitly the proteins and metabolites that mediate those interactions. A gene network is then a projection of the whole biochemical network onto a space where the only observables are gene transcripts (mRNA), but where the influence of the remaining biochemical system is felt implicitly (Brazhnik et al., 2002).

Gene networks are a logical way of attempting to describe phenomena observed with transcription profiling, such as is done with the popular microarray technology. Being able to create gene networks from experimental data and to use them to reason about their dynamics and design principles will contribute to increased understanding of cellular function.

Gene networks and microarray data

Our involvement in gene networks started back in 1999 when we were interested in finding out if clustering time series microarray data would really reveal "pathways". Our preliminary results (Mendes, 1999) show that this is not the case, at least with arbitrarily selected data.

More recently we have developed a system to construct gene network models and simulate experiments (including with added noise), which we call Artificial Gene Networks (Mendes et al., 2003). We have used this to study the effectiveness of single-gene t-tests, SAM, ANOVA, and mixture models, in identifying significant differences in gene expression in microarray data. These networks were also used in the 2nd DREAM competition.

Hierarchical control analysis

Metabolic control analysis is a well established methodology to study regulation of metabolic systems, but it can also be extended to include gene regulation (termed Hierarchical Control Analysis, or HCA for short). Using the principles of HCA we studied in which circumstances can one study just the metabolic part of the system without much error, i.e. neglect the gene regulatory component (de la Fuente et al., 2002a).

Reverse engineering gene networks

A holy grail of gene networks research is to be able to reconstruct interaction networks from functional data, such as from microarrays or RT-PCR. We have developed a method? to reconstruct the underlying gene regulatory network from microarray data obtained in carefully designed perturbation experiements (de la Fuente et al, 2001, 2002b, de la Fuente & Mendes, 2002). This method is based on co-control analysis and requires finite perturbations of gene expression rates.

A more established way to look at gene networks is based on correlation between levels of gene expression, be it by clustering or through Bayesian inference. We have proposed a new way to analyze gene expression data which attempts to eliminate the amount of correlation that is due to secondary effects (de la Fuente et al., 2004). This is based on partial correlation coefficients, which measure how much correlation exists between two variables (genes) when conditioned to a set of other genes. We suggested to use up to second order partial correlation coefficients to identify correlations that are not generated by direct interactions. The method is somewhat similar to Bayesian networks, though it is distinct in two major aspects: a) it is able to create cyclic dependency graphs, whereas Bayesian networks can only generate acyclic graphs; b) it is not able to assign directionality to interactions, whereas Bayesian networks are directed.

References

• Brazhnik, P., de la Fuente, A. & Mendes, P. (2002) Gene networks: how to put the function in genomics. Trends in Biotechnology 20, 467-472. [ abstract ] [ full text ]
• de la Fuente, A. & Mendes, P. (2002) Quantifying gene networks with regulatory strengths. Molecular Biology Reports 29, 73-77. [ abstract ]
• de la Fuente, A., Brazhnik, P. & Mendes, P. (2001). A quantitative method for reverse engineering gene networks from microarray experiments using regulatory strengths. Proceedings of the 2nd International Conference on Systems Biology, California Institute of Technology, Pasadena, CA. [ full text ] [ supplementary information ]
• de la Fuente, A., Snoep, J.L., Westerhoff, H.V. & Mendes, P. (2002a) Metabolic control in integrated biochemical systems. European Journal of Biochemistry 269, 4399-4408. [ abstract ] [ supplementary information ]
• de la Fuente, A., Brazhnik, P. & Mendes, P. (2002b) Linking the genes: inferring quantitative gene networks from microarray data. Trends in Genetics 18, 395-398. [ abstract ] [ supplementary information ]
• de la Fuente, A. & Mendes, P. (2003) Integrative modeling of gene expression and cell metabolism. Applied Bioinformatics 2, 79-90.
• Mendes, P. (1999) Metabolic simulation as an aid in understanding gene expression data. In Workshop on Computation of Biochemical Pathways and Genetic Networks (Bornberg-Bauer, E., De Beuckelaer, A., Kummer, U., and Rost, U., eds), Logos-Verlag, Heidelberg, pp. 27-34.
• Mendes, P., Sha, W. & Ye, K. (2003) Artificial gene networks for objective comparison of analysis algorithms. Bioinformatics 19, ii122-ii129. [ abstract ] [ full text ]