Differential Producibility Analysis (DPA) of transcriptomic data to help understand metabolic status of a cell or microorganism
While a vast amount of microarray data is now available freely at the various repositories at NCBI (GEO database) and EBI (ArrayExpress), there is currently no microarray data analysis methods available which can extract the global metabolic state of an microorganism or cell in vivo. Current methods only investigate the transcriptional activity of all encoded genes or identify differentially expressed genes in numerous microarray studies.
We recently developed a novel methodology ‘Differential producibility analysis (DPA)’ which will be appearing the Plos Computational Biology (Bonde et.al, Differential Producibility Analysis (DPA) of transcriptomic data with metabolic networks: Deconstructing the metabolic response of M. tuberculosis, 2011, PloS Comp. Biology, accepted, under press). I would like to provide the details of this method, its application and limitation and would like to present the critical unbiased view on the method.
What is DPA?
DPA is a meta-analysis method which allows identification of significantly differentially produced metabolites from the metabolic network and microarray data analysis. In a simple case DPA is a merger of two independent methods, Flux Balance analysis (1) of Genome Scale Metabolic Network models and modified Rank Product analysis (2). Hence DPA provides a snapshot of what’s going on in the metabolic pathway in terms of metabolites and helps in accelerating the biological understanding of the cell or microorganism’s two states (e.g wild-type vs mutant) in question.
Figure 1 explains the schematics of the DPA algorithm. I guess the figure is self explanatory in it-self, though an in detail description of algorithm can be found in the publication.
When DPA can be applied?
As long as you have a genome scale metabolic network available for your cell/tissue or microorganism which you can use to perform a Flux Balance Analysis and in silico gene knockout on the GSMN model, and you have at least two replicates of microarray experiment data for each state you want to analyse, you are ready to go. DPA was successfully applied to analyse large microarray data from M. Tuberculosis and E.coli and shown that DPA revealed
signature metabolites which were previously hypothesised to be over expressed or significantly produced in the given conditions of the two species.
DPA code is written in R and uses RankProd package from Bioconductor. The code is attached as a supplementary material, and at present, we are testing the DPA R package which will be available from Bioconductor repository.
DPA and other similar tools
There are few tools which can provide similar analysis, though the underlying algorithm is different. At a simple case, the GSEA algorithm (3) gives gene set enrichment analysis of predefined set of genes, while in DPA, gene sets are defined on the basis of those genes which affect the production of the metabolite (simulated using FBA approach, hence providing metabolite gene set analysis.). ‘Reporter metabolite’ by Patil et.al (4) performs similar meta-analysis, though the metabolite connectivity is taken from metabolic network and hence only based on local connectivity analysis than global metabolic network analysis.
1) FBA: Flux Balance analysis: Fell and Small, Fat synthesis in adipose tissue: an examination of stoichiometric constraints, Biochem J. 1986 September 15; 238(3): 781–786.
2) RankProd: Hong et. al. RankProd: a bioconductor package for detecting differentially expressed genes in meta-analysis, Bioinformatics, 22 (22), 2825-2827.
3) Subramanian, Tamayo, et al. (2005, PNAS 102, 15545-15550) and Mootha, Lindgren, et al. (2003, Nat Genet 34, 267-273).
4) Patil KR, Nielsen J (2005) Uncovering transcriptional regulation of metabolism using metabolic network topology. Proc Nat Acad Sci USA 102: 2685–2689