All publicly available annotated genomes within the clade Neocallimastigomycota were downloaded from the Joint Genome Institute’s (JGI) MycoCosm database (
48). This includes the high-quality PacBio-sequenced genomes of
Anaeromyces robustus,
Piromyces finnis, and
Neocallimastix californiae (
4) as well as the novel isolate
Neocallimastix lanati introduced here. The genomes of
Pecoramyces ruminatium, also known as
Orpinomyces sp. strain C1A (
7,
52), and
Piromyces sp. strain E2 (
4) were also included for completeness. The gene annotation data supplied by the JGI was combined with annotations derived from bidirectionally searching by blast (using BLASTp [
53]) the predicted genes from the gut fungal genomes against the curated Swiss-Prot database from UniProt (
54). Briefly, bidirectional blast searching annotates a predicted gut fungal gene if (i) the top hit using the fungal genome as the query and the reference collection as the database is the same as when (ii) the gut fungal genome is used as the database and the reference collection is used as the query. Furthermore, only matches with E values smaller than 1e
−20 were considered for assigning Enzyme Commission (EC) annotations to genes. This information was collated into a master metabolic table (see data set S3 in the iNlan20 GitHub repository available at
https://github.com/stelmo/iNlan20) and subsequently used to construct the model and assign genes to reactions. Enzyme complexes were assigned by using the “Subunit structure” field in the UniProt database. Protein localization was predicted using DeepLoc (
55). Reaction directions were primarily inferred from MetaCyc (
56), and specific Gibbs free energy change of reactions reported were calculated using eQuilibrator (
57). Transcriptomic and expression experiments for
N. lanati were conducted as part of this study (described later). These omics data sets were used to assign a confidence score to each gene in the model of
N. lanati. Gaps in the model of
N. lanati were filled by inspecting the EC assignments found for each other anaerobic fungus, as well as the GEMs of
E. coli and
S. cerevisiae, using the approach described above and looking for homologous genes in the genome of
N. lanati (
46,
58). The universal reactions and metabolites from the BiGG Models platform (
59) was used to construct the
in silico model where possible; if a reaction did not exist in that database it was manually added. The KEGG and MetaCyc databases were used as references to reconstruct the draft metabolic model based on the EC assignments of the metabolic annotation data (
56,
60). The curated model for
N. lanati was constructed by carefully following established genome-scale metabolic model construction protocols to refine the draft model (
26). Specifically, each reaction was inspected to ensure consistency, mass, and charge balance where possible. Model quality was benchmarked by the Memote application (see data set S1 in the iNlan20 GitHub repository at the above-mentioned URL) (
61). The curated
N. lanati model as well as the entire reconstruction pipeline and all the data used in this work can be found in the model repository at
https://github.com/stelmo/iNlan20. An experimentally measured flux of 1.5 mmol/g (dry weight)/h of glucose was used in all simulations. Flux balance analysis was used to simulate the genome-scale metabolic model of
N. lanati using the COBRA Toolbox and COBRApy (
62,
63). Flux samples (
N = 2,000) were generated by sampling from the model and constraining the objective function to be within 90% of the optimum found by FBA. This threshold was set to reflect the assumption that the gut fungi need to maintain a high growth rate to compete with faster growing bacteria in their native microbiome (
1). Escher was used to visualize the metabolism (
64). Example code that can be used to run the model and computational experiments is supplied as an IPython notebook in the model repository available at the URL mentioned above.