Inferring the role of transcription factors in regulatory networks
Abstract
Background: Expression profiles obtained from multiple perturbation experiments are increasingly used to reconstruct transcriptional regulatory networks, from well studied, simple organisms up to higher eukaryotes. Admittedly, a key ingredient in developing a reconstruction method is its ability to integrate heterogeneous sources of information, as well as to comply with practical observability issues: measurements can be scarce or noisy. The purpose of this work is (1) to build a formal model of regulations among genes; (2) to check its consistency with gene expression data on stress perturbation assays; (3) to infer the regulatory role of transcription factors as inducer or repressor if the model is consistent with expression profiles; (4) to isolate ambiguous pieces of information if it is not.
Results: We validate our methods on {\em E. Coli} network with a compendium of expression profiles. We investigate the dependence between the number of available expression profiles and the number of inferred regulations, in the case where all genes are observed. This is done by simulating artificial observations for the transcriptional network of {\em E. Coli} (1529 nodes and 3802 edges). We prove that at most 40,8\% of the network can be inferred and that 30 distinct expression profiles are enough to infer 30\% of the network on average. We repeat this procedure in the case of missing observations, and show that our approach is robust to a significant proportion of unobserved genes. Finally, we apply our inference algorithms to {\em S. Cerevisiae } transcriptional network, and demonstrate that for small scale subnetworks of {\em S. Cerevisiae} we are able to infer more than 20\% of the regulations. For more complex networks, we are able to detect and isolate inconsistencies between experimental sources and a non negligible portion of the model (15\% of all interactions).
Conclusions: Our approach does not require accurate expression levels, nor times series. Nevertheless, we show both on real and artificial data that a relatively small number of perturbation experiments are enough to determine a significant portion of regulatory effects. This is a key practical asset compared to statistical methods for network reconstruction. In addition, we illustrate the capability of our method to validate networks. We conjecture that inconsistencies we detected might be good candidates for further experimental investigations.
Results: We validate our methods on {\em E. Coli} network with a compendium of expression profiles. We investigate the dependence between the number of available expression profiles and the number of inferred regulations, in the case where all genes are observed. This is done by simulating artificial observations for the transcriptional network of {\em E. Coli} (1529 nodes and 3802 edges). We prove that at most 40,8\% of the network can be inferred and that 30 distinct expression profiles are enough to infer 30\% of the network on average. We repeat this procedure in the case of missing observations, and show that our approach is robust to a significant proportion of unobserved genes. Finally, we apply our inference algorithms to {\em S. Cerevisiae } transcriptional network, and demonstrate that for small scale subnetworks of {\em S. Cerevisiae} we are able to infer more than 20\% of the regulations. For more complex networks, we are able to detect and isolate inconsistencies between experimental sources and a non negligible portion of the model (15\% of all interactions).
Conclusions: Our approach does not require accurate expression levels, nor times series. Nevertheless, we show both on real and artificial data that a relatively small number of perturbation experiments are enough to determine a significant portion of regulatory effects. This is a key practical asset compared to statistical methods for network reconstruction. In addition, we illustrate the capability of our method to validate networks. We conjecture that inconsistencies we detected might be good candidates for further experimental investigations.
Origin | Files produced by the author(s) |
---|
Loading...