Methods

Construction of the M. alpina GSMM
The draft model was constructed by amassing reactions from the genome-scale metabolic models of genetically related organisms Aspergillus terreus [44], Pichia pastoris [45] and Yarrowia lipolytica [20]. Reactions were chosen based on orthologs shared between M. alpina and the three reference organisms identified by protein sequence similarity searches using BLAST. Protein sequences from A. terreus NIH 2624, P. pastoris GS 115 and Y. lipolytica CLIB 122 were downloaded from UniProt [46]. Open reading frame (ORF) information for M. alpina ATCC 32222 was provided by Yong Q. Chen based on sequencing results. The iJL1454 GSMM was used as a reference. Additionally, iLC915 [45] and iYL619_PCP [20] were also used for comparison since both models were concerned with lipid metabolism and overproduction. To ensure accuracy, only sufficiently similar orthologs with e-values ≤10−30 and sequence identities ≥40% were included [47]. To expand and update the draft model, the genome of M. alpina was re-annotated by submitting ORFs to the KAAS online annotation server [48]. Metabolic reactions absent from the draft model were added from the KEGG database [49] based on KAAS annotation results. MetaCyc [50] and BioPath [51] databases were used to judge reaction reversibility. Compartmentalization information assigned to reactions was determined by subcellular localization prediction tools CELLO [52] and WoLF PSORT. [53] BaCelLo [54] was also used for proteins that were difficult to determine with the other tools. Transport information was obtained by cross-referencing BLATSp searches and the Transporter Classification Database TCDB [55].
To refine the draft model, the gapFind [56] program in the COBRA Toolbox [22] was used to identify metabolic gaps in draft model and literature data were used to fill these gaps. The metabolites in each reaction were characterized based on their chemical formulae and neutral charges, which were obtained using CHEBI [57] and PubChem [58].

Biomass composition
The biomass equation of M. alpina was assumed to have six components: proteins, DNA, RNA, lipids, the cell wall and the small molecule pool [59,60]. Since no detailed information on M. alpina DNA and RNA was available, the ratio was assumed to be the same as in the related Aspergillus niger [61]. The nucleotide and amino acid composition were calculated based on the M. alpina ATCC 32222 genome [2], as no specific experimental data were available. Similarly, the cell wall composition was calculated based on the typical fungal cell wall structure [62]. The lipid composition was calculated based on the current literature [63]. For calculation of energetic parameters, the growth and non-growth associated ATP maintenance values (GAM and NGAM, respectively) were assumed to be the same as those in the central carbon metabolic model of A. niger [64]. Detailed biomass composition information can be found in Additional file 7.

Simulation and analysis
The reconstructed metabolic network was converted into stoichiometric matrix (S = M * N) using the Matlab program, where M represents metabolites and N represents reactions [65]. The basic tools used for model analysis were flux balance analysis (FBA) and flux variability analysis (FVA). GLPK was used for linear programming [22], and Gurobi was used for quadratic programming [66]. In silico analysis included growth simulation, gene and reaction essentiality analysis, robustness analysis, and minimization of metabolic adjustment. Analyses were performed according to the instructions for the COBRA Toolbox [22]. Constraints used in each analysis are mentioned in the results section. To analyze model parameters relevant to cell growth, the biomass equation was selected as the objective function. For analysis of ARA production, the exchange reaction of ARA was the objective function.