Limits...
Refining transcriptional regulatory networks using network evolutionary models and gene histories.

Zhang X, Moret BM - Algorithms Mol Biol (2010)

Bottom Line: In previous work, we used a simple evolutionary model and provided extensive simulation results showing that phylogenetic information, combined with such a model, could be used to gain significant improvements on the performance of current inference algorithms.We show how to adapt our evolutionary approach to this new model and provide detailed simulation results, which show significant improvement on the reference network inference algorithms.We also provide results on biological data (cis-regulatory modules for 12 species of Drosophila), confirming our simulation results.

View Article: PubMed Central - HTML - PubMed

Affiliation: Laboratory for Computational Biology and Bioinformatics, Ecole Polytechnique Fédérale de Lausanne, EPFL-IC-LCBB, INJ230, Station 14, CH-1015 Lausanne, Switzerland.

ABSTRACT

Background: Computational inference of transcriptional regulatory networks remains a challenging problem, in part due to the lack of strong network models. In this paper we present evolutionary approaches to improve the inference of regulatory networks for a family of organisms by developing an evolutionary model for these networks and taking advantage of established phylogenetic relationships among these organisms. In previous work, we used a simple evolutionary model and provided extensive simulation results showing that phylogenetic information, combined with such a model, could be used to gain significant improvements on the performance of current inference algorithms.

Results: In this paper, we extend the evolutionary model so as to take into account gene duplications and losses, which are viewed as major drivers in the evolution of regulatory networks. We show how to adapt our evolutionary approach to this new model and provide detailed simulation results, which show significant improvement on the reference network inference algorithms. Different evolutionary histories for gene duplications and losses are studied, showing that our adapted approach is feasible under a broad range of conditions. We also provide results on biological data (cis-regulatory modules for 12 species of Drosophila), confirming our simulation results.

No MeSH data available.


Performance of refinement algorithms on Drosophila data, with basic network evolution model. (A) Results on CRM abd-A_iab-2_1.7_; (B) Results on CRM Abd-B_IAB5.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2823753&req=5

Figure 1: Performance of refinement algorithms on Drosophila data, with basic network evolution model. (A) Results on CRM abd-A_iab-2_1.7_; (B) Results on CRM Abd-B_IAB5.

Mentions: We conducted experiments on different CRMs of the 12 Drosophila species; here we show results on two of them. In both experiments, regulatory networks have 6 TFs and 12 target genes, forming networks with 18 nodes. Average performance for the base inference algorithm (DBI) and for the two refinement algorithms over 10 runs for these two experiments is shown in Fig. 1 using ROC curves. In the two plots, the points on each curve are obtained with different structure complexity penalty coefficients. From Fig. 1 we can see the improvement of our refinement algorithms over the base algorithm is significant: RefineML improves significantly both sensitivity and specificity, while RefineFast loses a little sensitivity while gaining more specificity for sparse networks. (In both CRMs, the standard deviation on sensitivity is around 0.05 and that on specificity around 0.005.) The dominance of RefineML over RefineFast shows the advantage of reusing the leaf networks inferred by base algorithms, especially when the error rate in these leaf networks is low. Results on other CRMs show similar improvement of our refinement algorithms. Besides the obvious improvement, we can also observe the fluctuation of the curves: theoretically sensitivity can be traded for specificity and vice versa, so that the ROC curves should be in "smooth" shapes, which is not the case for Fig. 1. Various factors can account for this: the shortage of gene-expression data to infer the network, the noise inherent in biological data, the special structure of the networks to be inferred, or the relatively small amount of data involved, leading to higher variability. (We have excluded the first possibility by generating larger gene-expression datasets for inference algorithms, where similar fluctuations still occur.)


Refining transcriptional regulatory networks using network evolutionary models and gene histories.

Zhang X, Moret BM - Algorithms Mol Biol (2010)

Performance of refinement algorithms on Drosophila data, with basic network evolution model. (A) Results on CRM abd-A_iab-2_1.7_; (B) Results on CRM Abd-B_IAB5.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2823753&req=5

Figure 1: Performance of refinement algorithms on Drosophila data, with basic network evolution model. (A) Results on CRM abd-A_iab-2_1.7_; (B) Results on CRM Abd-B_IAB5.
Mentions: We conducted experiments on different CRMs of the 12 Drosophila species; here we show results on two of them. In both experiments, regulatory networks have 6 TFs and 12 target genes, forming networks with 18 nodes. Average performance for the base inference algorithm (DBI) and for the two refinement algorithms over 10 runs for these two experiments is shown in Fig. 1 using ROC curves. In the two plots, the points on each curve are obtained with different structure complexity penalty coefficients. From Fig. 1 we can see the improvement of our refinement algorithms over the base algorithm is significant: RefineML improves significantly both sensitivity and specificity, while RefineFast loses a little sensitivity while gaining more specificity for sparse networks. (In both CRMs, the standard deviation on sensitivity is around 0.05 and that on specificity around 0.005.) The dominance of RefineML over RefineFast shows the advantage of reusing the leaf networks inferred by base algorithms, especially when the error rate in these leaf networks is low. Results on other CRMs show similar improvement of our refinement algorithms. Besides the obvious improvement, we can also observe the fluctuation of the curves: theoretically sensitivity can be traded for specificity and vice versa, so that the ROC curves should be in "smooth" shapes, which is not the case for Fig. 1. Various factors can account for this: the shortage of gene-expression data to infer the network, the noise inherent in biological data, the special structure of the networks to be inferred, or the relatively small amount of data involved, leading to higher variability. (We have excluded the first possibility by generating larger gene-expression datasets for inference algorithms, where similar fluctuations still occur.)

Bottom Line: In previous work, we used a simple evolutionary model and provided extensive simulation results showing that phylogenetic information, combined with such a model, could be used to gain significant improvements on the performance of current inference algorithms.We show how to adapt our evolutionary approach to this new model and provide detailed simulation results, which show significant improvement on the reference network inference algorithms.We also provide results on biological data (cis-regulatory modules for 12 species of Drosophila), confirming our simulation results.

View Article: PubMed Central - HTML - PubMed

Affiliation: Laboratory for Computational Biology and Bioinformatics, Ecole Polytechnique Fédérale de Lausanne, EPFL-IC-LCBB, INJ230, Station 14, CH-1015 Lausanne, Switzerland.

ABSTRACT

Background: Computational inference of transcriptional regulatory networks remains a challenging problem, in part due to the lack of strong network models. In this paper we present evolutionary approaches to improve the inference of regulatory networks for a family of organisms by developing an evolutionary model for these networks and taking advantage of established phylogenetic relationships among these organisms. In previous work, we used a simple evolutionary model and provided extensive simulation results showing that phylogenetic information, combined with such a model, could be used to gain significant improvements on the performance of current inference algorithms.

Results: In this paper, we extend the evolutionary model so as to take into account gene duplications and losses, which are viewed as major drivers in the evolution of regulatory networks. We show how to adapt our evolutionary approach to this new model and provide detailed simulation results, which show significant improvement on the reference network inference algorithms. Different evolutionary histories for gene duplications and losses are studied, showing that our adapted approach is feasible under a broad range of conditions. We also provide results on biological data (cis-regulatory modules for 12 species of Drosophila), confirming our simulation results.

No MeSH data available.