Reconstruction of the GEMs was comprised by a semi-automatic approach enriched with functional analysis and extensive manual curation based on available literature. Figure 1 depicts an overview of the process. The RAVEN Toolbox was used to generate draft models (manuscript under review). This approach uses a set of reference GEMs of closely related species and infers reactions by means of protein homology. In parallel to this method, the RAVEN Toolbox also uses the KEGG database to infer reactions that may be missing or incorrect in the template models. Draft models were constructed using the genome sequences of P. stipitis (CBS 6054) [20 (link)], and P. pastoris (GS 115) [19 (link)]. The iIN800 GEM of S. cerevisiae was used as reference framework because of its accurate annotation of fatty acid metabolism, and the extensive information about metabolites and genes, which allowed for a convenient comparison with the KEGG database.
BLASTp (ncbi-blast software ver. 2.2.24) was used to identify homologous proteins among the three yeast species. Protein homologs were identified based on stringent cutoff values (E-values < 10-40), and on the score to sequence length ratio according to David et al. (2008) [31 (link)]. KEGG Ontology (KO) identifiers were also used to additionally infer reactions which could not be found in S. cerevisiae from the genome sequences of the two Pichia species following the RAVEN Toolbox pipeline. Finally, the metabolic network of S. cerevisiae iIN800 was used to map genes from P. pastoris and P. stipitis having homologs in S. cerevisiae.
Subcellular compartmentalization of reactions was determined using the F-LocA (Fully-connected Localization Assignment), which is part of the RAVEN Toolbox. F-LocA incorporates subcellular localization predictors (CELLO and WoLFPSORT) [32 (link)], together with a constraint on network connectivity. Reactions without associated genes were compartmentalized according to biochemical evidence when available. It is important to note that these automated approaches were only used as an aid in the reconstructions, and that biochemical and physiological evidence was always used to validate reaction localizations and gene associations. This was of particular importance in the peroxisomal metabolism where the predictive capability is lower due to the low quality of data from subcellular localization predictors (e.g. CELLO predicts that AOX is in the cytosol, but it is in the peroxisome). In cases where information about P. pastoris or P. stipitis was lacking, data from other closely related yeasts was used instead (e.g. S. cerevisiae, Hansenula polymorpha, Candida tropicalis, C. shehatea, and C. boidinii [33 (link),34 (link)]). Both GEMs are available in the BioMet Toolbox [http://www.sysbio.se/BioMet/ - will be uploaded upon acceptance of paper].
BLASTp (ncbi-blast software ver. 2.2.24) was used to identify homologous proteins among the three yeast species. Protein homologs were identified based on stringent cutoff values (E-values < 10-40), and on the score to sequence length ratio according to David et al. (2008) [31 (link)]. KEGG Ontology (KO) identifiers were also used to additionally infer reactions which could not be found in S. cerevisiae from the genome sequences of the two Pichia species following the RAVEN Toolbox pipeline. Finally, the metabolic network of S. cerevisiae iIN800 was used to map genes from P. pastoris and P. stipitis having homologs in S. cerevisiae.
Subcellular compartmentalization of reactions was determined using the F-LocA (Fully-connected Localization Assignment), which is part of the RAVEN Toolbox. F-LocA incorporates subcellular localization predictors (CELLO and WoLFPSORT) [32 (link)], together with a constraint on network connectivity. Reactions without associated genes were compartmentalized according to biochemical evidence when available. It is important to note that these automated approaches were only used as an aid in the reconstructions, and that biochemical and physiological evidence was always used to validate reaction localizations and gene associations. This was of particular importance in the peroxisomal metabolism where the predictive capability is lower due to the low quality of data from subcellular localization predictors (e.g. CELLO predicts that AOX is in the cytosol, but it is in the peroxisome). In cases where information about P. pastoris or P. stipitis was lacking, data from other closely related yeasts was used instead (e.g. S. cerevisiae, Hansenula polymorpha, Candida tropicalis, C. shehatea, and C. boidinii [33 (link),34 (link)]). Both GEMs are available in the BioMet Toolbox [
Free full text: Click here
