Various approaches have been taken to analyse regulatory networks in the last few years. One such approach is in silico
simulation of fluctuations in network components when the network is perturbed [1
]. Comparing in silico
results with experimental outputs may highlight the relevance of certain approaches in our understanding of biological insights. Nonetheless, confronting experimental data with a regulatory network comes up against many difficulties. One of these is the incomplete state-of-art of the regulatory knowledge, recently identified in [2
]. Another difficulty is the large scale of regulatory models, which makes the task of collecting kinetic constants and parameters intolerable. Other issues that induce errors in this confrontation are heterogeneity in the available data [3
], and the sometimes violated assumption that mRNA expression levels are correlated to protein activity [4
]. To deal with these problems, we propose a global
qualitative analysis, which formalises automatic reasoning in order to compare experimental data with the network behaviour expected from its topology. Our method highlights incomplete regions in a regulatory model and reasons over which products in the network need to be activated or inactivated in order to globally
explain the experimental data [5
As described in this article, we have enclosed this reasoning in the BioQuali Cytoscape [8
] plugin, which allows the user to visualise the outputs of global automatic reasoning on large-scale regulatory networks. These outputs can be inconsistent subgraphs or a list of nodes, whose expression is deduced as increasing or decreasing in order to explain the observed data consistently.
Consistency between regulatory networks and expression data has been studied previously in [3
]. These studies proposed manual local consistency checks, in some cases by adding new regulation rules. The automation of this idea was recently implemented in a Cytoscape plugin in [10
]. With the BioQuali plugin, we go one step further since we propose automating a global consistency reasoning: the effects of influences are carried through the whole network in order to generate a consistency diagnosis that takes account of the entire topology of the network. In comparison with previous approaches, the new functions we provide are: a global analysis, which involves all network connections; no assumption regarding the null transcription factor activity when its mRNA expression change is non-significant; and visualisation of a consistent configuration of the whole network in which it is inferred that the expression of certain nodes fluctuates in a determined way that explains the experimental data provided. To do this, the BioQuali plugin relies on a robust architecture, which makes it effective in terms of speed and maintainability. BioQuali uses statistically significant expression profiles to deduce activity levels of transcription factors (TFs), however, it does not assume that non-significant mRNA changes imply a null activity level of the TF. The BioQuali plugin can help detect post-transcriptional regulations in a model; this capacity relies on the BioQuali inferred fluctuating nodes. This inference, obtained after confronting consistent transcriptional regulations with mRNA changes, reflects new fluctuations related either to mRNA expression or to protein activity levels. Confronting this inference with the sense of variation of mRNA expression may elucidate missing post-transcriptional interactions. Detecting an inconsistent subgraph may also imply the absence of a post-transcriptional regulation. BioQuali cannot automatically predict when a post-transcriptional interaction is missing, but it does provide tools for diagnosing a regulatory network which, combined with biological expertise, may result in adding post-transcriptional effects to complete a network model.
We illustrate this plugin by means of two regulatory models: a small eukaryotic model of the regulation of fatty acid metabolism and the large-scale prokaryotic transcriptional network of E. coli. These networks have been compared with small- and large-scale datasets, obtaining in each case globally-explained inconsistencies and predictions that agree with the biological literature.