Large-scale dynamic gene regulatory network inference combining differential equation models with local dynamic Bayesian network analysis. Bioinformatics

Monsanto Company, Mail zone CC1A, Chesterfield, MO 63017, USA.
Bioinformatics (Impact Factor: 4.98). 08/2011; 27(19):2686-91. DOI: 10.1093/bioinformatics/btr454
Source: PubMed


Reverse engineering gene regulatory networks, especially large size networks from time series gene expression data, remain a challenge to the systems biology community. In this article, a new hybrid algorithm integrating ordinary differential equation models with dynamic Bayesian network analysis, called Differential Equation-based Local Dynamic Bayesian Network (DELDBN), was proposed and implemented for gene regulatory network inference.
The performance of DELDBN was benchmarked with an in vivo dataset from yeast. DELDBN significantly improved the accuracy and sensitivity of network inference compared with other approaches. The local causal discovery algorithm implemented in DELDBN also reduced the complexity of the network inference algorithm and improved its scalability to infer larger networks. We have demonstrated the applicability of the approach to a network containing thousands of genes with a dataset from human HeLa cell time series experiments. The local network around BRCA1 was particularly investigated and validated with independent published studies. BRAC1 network was significantly enriched with the known BRCA1-relevant interactions, indicating that DELDBN can effectively infer large size gene regulatory network from time series data.
The R scripts are provided in File 3 in Supplementary Material.;
Supplementary data are available at Bioinformatics online.

Download full-text


Available from: Zheng Li, Oct 07, 2014
  • Source
    • "For the machine learning-based approaches, the network is inferred through measuring the dependences or causalities between transcriptional factors (TFs) and target genes (Ku¨ffner et al., 2012). Popular methods in this category include partial correlation coefficient (De la Fuente et al., 2004, Saito et al., 2011), Bayesian network analysis (Li et al., 2011; Yeung et al., 2011), mutual information (MI) (Basso et al., 2005; Belcastro et al., 2011; Faith et al., 2007; Margolin et al., 2006b; Modi et al., 2011) and conditional mutual information (CMI) (Sumazin et al., 2011; Zhang et al., 2012). As one of the most popular methods, MI has been widely used to construct GRNs because it provides a natural generalization of correlation owing to its capability of characterizing non-linear dependency (Brunel et al., 2010). "

  • Source
    • "These methods are deterministic rather than statistical, although ODE methods can be combined with statistical methods. DBN on local networks within a larger ODE model inference method have been used, for example [8]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Genome-wide time-series data provide a rich set of information for discovering gene regulatory relationships.As genome-wide data for mammalian systems are being generated, it is critical to developnetwork inference methods that can handle tens of thousands of genes efficiently, provide a systematicframework for the integration of multiple data sources, yield robust, accurate and compact gene-togenerelationships. We developed and applied ScanBMA, a Bayesian inference method that incorporates external informationto improve the accuracy of the inferred network. In particular, we developed a new strategy toefficiently search the model space, applied data transformations to reduce the effect of spurious relationships,and adopted the g-prior to guide the search for candidate regulators. Our method is highlycomputationally efficient, thus addressing the scalability issue with network inference. The method isimplemented as the ScanBMA function in the networkBMA Bioconductor software package. We compared ScanBMA to other popular methods using time series yeast data as well as time-seriessimulated data from the DREAM competition. We found that ScanBMA produced more compact networkswith a greater proportion of true positives than the competing methods. Specifically, ScanBMAgenerally produced more favorable areas under the Receiver-Operating Characteristic and Precision-Recall curves than other regression-based methods and mutual-information based methods. In addition,ScanBMA is competitive with other network inference methods in terms of running time.
    BMC Systems Biology 04/2014; 8(1):47. DOI:10.1186/1752-0509-8-47 · 2.44 Impact Factor
  • Source
    • "[20], but when time intervals are long (e.g., in hours), this approximation is very inaccurate. [20,27] suggested to employ a linear first order Markov model, which assumed the expression of genes at time k, as a linear function of its regulators at the previous time k – 1, i.e., "
    [Show abstract] [Hide abstract]
    ABSTRACT: During embryogenesis, signaling molecules produced by one cell population direct gene regulatory changes in neighboring cells and influence their developmental fates and spatial organization. One of the earliest events in the development of the vertebrate embryo is the establishment of three germ layers, consisting of the ectoderm, mesoderm and endoderm. Attempts to measure gene expression in vivo in different germ layers and cell types are typically complicated by the heterogeneity of cell types within biological samples (i.e., embryos), as the responses of individual cell types are intermingled into an aggregate observation of heterogeneous cell types. Here, we propose a novel method to elucidate gene regulatory circuits from these aggregate measurements in embryos of the frog Xenopus tropicalis using gene network inference algorithms and then test the ability of the inferred networks to predict spatial gene expression patterns. We use two inference models with different underlying assumptions that incorporate existing network information, an ODE model for steady-state data and a Markov model for time series data, and contrast the performance of the two models. We apply our method to both control and knockdown embryos at multiple time points to reconstruct the core mesoderm and endoderm regulatory circuits. Those inferred networks are then used in combination with known dorsal-ventral spatial expression patterns of a subset of genes to predict spatial expression patterns for other genes. Both models are able to predict spatial expression patterns for some of the core mesoderm and endoderm genes, but interestingly of different gene subsets, suggesting that neither model is sufficient to recapitulate all of the spatial patterns, yet they are complementary for the patterns that they do capture. The presented methodology of gene network inference combined with spatial pattern prediction provides an additional layer of validation to elucidate the regulatory circuits controlling the spatial-temporal dynamics in embryonic development.
    BMC Systems Biology 01/2014; 8(1):3. DOI:10.1186/1752-0509-8-3 · 2.44 Impact Factor
Show more