A community-driven, genome-scale metabolic reconstruction and analysis toolkit for Chinese Hamster Ovary (CHO) cells.
Quick links: Installation · Create model · Pre-process · Citing
Highlights
- Scope: 11,004 reactions • 7,377 metabolites • 3,597 genes • 3,489 mapped protein structures
- Use cases: context-specific modeling (WT vs ZeLa), pFBA/ecFBA, flux sampling, subsystem coverage, structure-guided hypotheses (e.g., putative PEP allosteric inhibition of PFK)
- Artifacts: curated datasets, notebooks (Python & MATLAB), figures, and utilities to reproduce key analyses
Model provided in this repository
Note: The model files distributed in
iCHO3K/Model/correspond to an unblocked version of the full iCHO3K network. In this version, blocked reactions have been removed and boundary reactions have been configured to support context-specific model generation with tools such as mCADRE.
The unblocked iCHO3K model contains 8,370 reactions, 4,695 metabolites, and 2,929 genes.
To generate the full iCHO3K model, follow the steps in Create iCHO3K Model.
If you use this repository or the iCHO3K model, please see Citing below.
- Repository layout
- Installation & setup
- Create iCHO3K Model
- Pre-process iCHO3K for context-specific model generation
- Reproducing key analyses
- Data & provenance
- Model formats & I/O
- Solvers & performance tips
- Contributing
- Citing
- Maintainers & contact
- Acknowledgments
- FAQ / Troubleshooting
Analyses/
├── conf_score_distribution.png # Confidence Score distribution across all reactions from iCHO3K
├── data_preprocessing/ # ZeLa vs WT growth rate and spent media data analysis
├── ecFBA and Flux Sum/ # ZeLa vs WT ecFBA and Flux Sum datasets
├── growth_rate_pred/ # pFBA simulations from ZeLa and WT context-specific models
├── recons_comparisons/ # Plot comparisions between iCHO3K and previous CHO reconstructions
├── Relevant_mets/ # Analysis of subsystems related to metabolites relevant to biomass sysnthesis
├── sampled_results/
├── subsystem_overview/ #Subsystem/System classification sunburst plot
└── tSNE/ #tSNE embedding analysis
Data/
├── Context_specific_models/ # Context-specific ZeLa and WT reduced and ec models (MAT)
│ └── unblocked_ecModel_generic/ # Generic iCHO3K ec model
│
├── GPR_Mapping/ # Supplementary data for GPR Mapping from Recon3D to iCHO3K
├── Gene_Essentiality/ # Set of experimentally validated CHO essential genes
├── Metabolites/ # Supplementary data for metabolites information
├── Orthologs/ # Ortholog mapping information from Human to Chinese Hamster
├── Reconciliation/ # Source reconstructions & derived models and datasets
│ ├── datasets/
│ └── models/
│
├── Uptake_Secretion_Rates/ # Pre-processed uptake and secretion rates from ZeLa fed-batch data
└── ZeLa Data/ # ZeLa 14 fed-batch raw transcriptomics and spent media data
iCHO3K/
├── Dataset/ # iCHO3K source dataset for model generation
└── Model/ # iCHO3K generic **unblocked** model variants
Matlab/
├── ecFBA/ # ecFBA scripts
├── Model_Extraction/ # Model Extraction with mCADRE scripts
├── Standep/ # ubiquityScore calculation with Standep
├── main_Model_Extraction.m # Main code for mCADRE model extraction
└── main_standep.m # Main code for Standep data preprocessing
Python/
├── Network Reconstruction/ # Notebooks related to the reconciliation of previous reconstructions and building of iCHO3K
│ ├── Genes.ipynb # Retrieval of Gene information from databases
│ ├── Metabolites.ipynb # Integration of metabolite information, de-duplication and analysis
│ ├── Reactions.ipynb # Reconciliation of previous CHO and Recon3D reconstructions, de-duplication, subsytem re-organization
│ └── retrieveTurnoverNumber.ipynb # Fetch turnover numbers and molecular weights from the BRENDA
│
├── Supplementary Notebooks/ # Supplementary Notebooks with extra information of previous reconstructions
├── Comparison..Reconstructions.ipynb # Comparison of iCHO3K with previous CHO reconstructions
├── Computational_Tests.ipynb # pFBA Simulations & tSNE embeddings
├── Model_builder.ipynb # iCHO3K Model Buider
├── Calculate_specific_rates.ipynb # Preprocess of spent media data into GEM fluxes
└── ZeLa_fluxomics.ipynb # ZeLa fluxomics data
Large files: Some assets may use Git LFS. If you see pointer files, run:
git lfs install git lfs pull --include="Data/**"Optional: clone without data, then fetch only Data/:
GIT_LFS_SKIP_SMUDGE=1 git clone <repo-url> cd iCHO3K && git lfs install git lfs pull --include="Data/**" --exclude=""
Create the exact conda environment shipped with the repo:
conda env create -f environment.yml
conda activate icho3kIf you prefer pip, use the lightweight set (no native solvers or libSBML):
python -m venv .venv && source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements.txt# conda-forge (recommended)
conda install -c conda-forge graphviz pygraphviz ndex2
# If you use pip for pygraphviz, ensure system Graphviz is installed first.If you plan to use commercial solvers (e.g., Gurobi/CPLEX), install them separately and set the solver in your code (see Solvers & performance tips).
- MATLAB R2022b+ recommended (earlier versions likely workable).
- Add
Notebooks/Matlab/to your MATLAB path.
import pandas as pd
import cobra
from cobra import Model, Reaction, Metabolite
#Read iCHO3K Full Dataset (11,004 reactions • 7,377 metabolites • 3,597 genes)
FILE_PATH = 'iCHO3K/Dataset/iCHO3K.xlsx'
metabolites = pd.read_excel(FILE_PATH, sheet_name='Metabolites')
rxns = pd.read_excel(FILE_PATH, sheet_name='Rxns')
rxns_attributes = pd.read_excel(FILE_PATH, sheet_name='Attributes')
boundary_rxns = pd.read_excel(FILE_PATH, sheet_name='BoundaryRxns')
genes_df = pd.read_excel(FILE_PATH, sheet_name='Genes')
# Create a cobra model and add reactions
model = Model("iCHO3K_vX")
lr = []
for _, row in rxns.iterrows():
r = Reaction(row['Reaction'])
lr.append(r)
model.add_reactions(lr)
# Add reaction-specific information
for i,r in enumerate(tqdm(model.reactions)):
r.build_reaction_from_string(rxns['Reaction Formula'][i])
r.name = rxns['Reaction Name'][i]
r.subsystem = rxns['Subsystem'][i]
if not (pd.isna(rxns['GPR_iCHO3K'][i]) or rxns['GPR_iCHO3K'][i] == ''):
r.gene_reaction_rule = str(rxns['GPR_iCHO3K'][i])
r.lower_bound = float(rxns_attributes['Lower bound'][i])
r.upper_bound = float(rxns_attributes['Upper bound'][i])
r.annotation['confidence_score'] = str(rxns['Conf. Score'][i])
r.annotation['molwt'] = str(rxns_attributes['Mol wt'][i])
r.annotation['kcat_f'] = str(rxns_attributes['kcat_forward'][i])
r.annotation['kcat_b'] = str(rxns_attributes['kcat_backward'][i])
# Add Boundary Reactions
dr = []
for _, row in boundary_rxns.iterrows():
r = Reaction(row['Reaction'])
dr.append(r)
model.add_reactions(dr)
boundary_rxns_dict = boundary_rxns.set_index('Reaction').to_dict()
boundary_rxns_dict
for i,r in enumerate(tqdm(model.reactions)):
if r in dr:
r.build_reaction_from_string(boundary_rxns_dict['Reaction Formula'][r.id])
r.name = boundary_rxns_dict['Reaction Name'][r.id]
r.subsystem = boundary_rxns_dict['Subsystem'][r.id]
r.lower_bound = float(boundary_rxns_dict['Lower bound'][r.id])
r.upper_bound = float(boundary_rxns_dict['Upper bound'][r.id])
r.annotation['confidence_score'] = str(1)
# Add information for each metabolite
metabolites_dict = metabolites.set_index('BiGG ID').to_dict('dict')
for met in model.metabolites:
try:
met.name = metabolites_dict['Name'][f'{met}']
met.formula = metabolites_dict['Formula'][f'{met}']
met.compartment = metabolites_dict['Compartment'][f'{met}'].split(' - ')[0]
try:
met.charge = int(metabolites_dict['Charge'][f'{met}'])
except (ValueError, TypeError):
print(f'{met} doesnt have charge')
except (KeyError):
print('----------------------------')
print(f'{met} doesnt exist in the df')
print('----------------------------')
#OPTIONAL: Add Gene Name information
genes_dict = genes_df.iloc[:,:2].set_index('Gene Entrez ID').to_dict('dict')
for g in model.genes:
if g.id in list(genes_dict['Gene Symbol'].keys()):
g.name = genes_dict['Gene Symbol'][f'{g}']
# Set biomass objective function and save the model
name = 'CHO-Generic' # OPTIONS: 'CHO-S', 'CHO-Generic', 'CHO-Prod-Generic'
if name == 'CHO-Generic':
model.objective = 'biomass_cho'
elif name == 'CHO-S':
model.objective = 'biomass_cho_s'
elif name == 'CHO-Prod-Generic':
model.objective = 'biomass_cho_prod'
else:
raise ValueError(f"Unrecognized model name: {name}")
# OPTIONAL: remove infeasible loop-generating reactions:
glucloop_reactions = [
'GapFill-R01206', 'GAUGE-R00557',
'GAUGE-R00558', 'FNOR', 'GGH', 'r0741', 'r1479', 'XOLESTPOOL'
]
try:
model.remove_reactions(glucloop_reactions, remove_orphans=True)
except KeyError:
print(f'Reaction {reaction_id} not in model {model.id}')
atp_loop_reactions = [
'SCP22x','TMNDNCCOAtx','OCCOAtx','r0391','BiGGRxn67','r2247','r2280',
'r2246','r2279','r2245','r2305','r2317','r2335','HMR_0293','HMR_7741',
'r0509','r1453','HMR_4343','ACONTm','PDHm','r0426','r0383','r0555',
'r1393','NICRNS','GAUGE-R00648','GAUGE-R03326','GapFill-R08726','RE2915M',
'HMR_3288','HMR_1325','HMR_7599','r1431','r1433','RE2439C','r0791',
'r1450','GAUGE-R00270','GAUGE-R02285','GAUGE-R04283','GAUGE-R06127','GAUGE-R06128',
'GAUGE-R06238','GAUGE-R00524','RE3477C','AAPSAS','RE3347C','HMR_0960','HMR_0980',
'RE3476C','r0708','r0777','r0424','r0698','3HDH260p','HMR_3272','ACOAD183n3m',
'HMR_1996','GapFill-R01463','GapFill-R04807','r1468','r2435','r0655','r0603','r0541',
'RE0383C','HMR_1329','TYRA','NRPPHRt_2H','GAUGE-R07364','GapFill-R03599','ARD',
'RE3095C','RE3104C','RE3104R','ACONT','ICDHxm','ICDHy',
'r0425','r0556','NH4t4r','PROPAT4te','r0085','r0156','r0464','ABUTDm',
'OIVD1m','OIVD2m','OIVD3m','r2194','r2202','HMR_9617','r2197','r2195',
'2OXOADOXm','r2328','r0386','r0451','FAS100COA','FAS120COA','FAS140COA',
'FAS80COA_L','r0604','r0670','r2334','r0193','r0595','r0795','GLYCLm',
'MACACI','r2193','r0779','r0669','UDCHOLt','r2146','r2139'
]
try:
model.remove_reactions(atp_loop_reactions, remove_orphans=True)
except KeyError:
print(f'Reaction {reaction_id} not in model {model.id}')
##### ----- Save iCHO3K full version (11,004 reactions • 7,377 metabolites • 3,597 genes) ----- #####
suffix = name.lower().replace('-', '_')
# XML
model_name_xml = f'../iCHO3K/Model/iCHO3K_{suffix}.xml'
write_sbml_model(model, model_name_xml)
# JSON, because the sbml doesnt save the subsystems
model_name_json = f'../iCHO3K/Model/iCHO3K_{suffix}.json'
save_json_model(model, model_name_json)
# MATLAB
model_name_matlab = f'../iCHO3K/Model/iCHO3K_{suffix}.mat'
save_matlab_model(model, model_name_matlab)from cobra.flux_analysis import find_blocked_reactions, flux_variability_analysis
## Step 1: Identify blocked reactions
for rxn in model.boundary:
if rxn.id.startswith("EX_"):
rxn.bounds = (-1000,1000)
if rxn.id.startswith("SK_"):
rxn.bounds = (-1000,1000)
if rxn.id.startswith("DM_"):
rxn.bounds = (0,1000)
model.solver = 'gurobi'
blocked_reactions = find_blocked_reactions(model)
## Step 2: Remove blocked reactions from the model
model_unblocked = model.copy()
blocked_reaction_objects = [model_unblocked.reactions.get_by_id(rxn_id) for rxn_id in blocked_reactions]
model_unblocked.remove_reactions(blocked_reaction_objects, remove_orphans=True)
print(f"Removed {len(blocked_reaction_objects)} blocked reactions from the model.")
## Step 3: Set the bounds for context-specific model generation
for rxn in model.boundary:
# Models that are forced to secrete ethanol are not feasible
if rxn.id == 'EX_etoh_e':
rxn.bounds = (-1,1)
continue
# Keep boundaries open for essential metabolites
if rxn.id == 'EX_h2o_e':
rxn.bounds = (-1000,1000)
continue
if rxn.id == 'EX_h_e':
rxn.bounds = (-1000,1000)
continue
if rxn.id == 'EX_o2_e':
rxn.bounds = (-1000,1000)
continue
if rxn.id == 'EX_hco3_e':
rxn.bounds = (-1000,1000)
continue
if rxn.id == 'EX_so4_e':
rxn.bounds = (-1000,1000)
continue
if rxn.id == 'EX_pi_e':
rxn.bounds = (-1000,1000)
continue
# Boundaries from Sink reactions on iCHO_v1 (100 times lower)
if rxn.id == 'SK_Asn_X_Ser_Thr_r':
rxn.bounds = (-0.001,1000)
continue
if rxn.id == 'SK_Tyr_ggn_c':
rxn.bounds = (-0.001,1000)
continue
if rxn.id == 'SK_Ser_Thr_g':
rxn.bounds = (-0.001,1000)
continue
if rxn.id == 'SK_pre_prot_r':
rxn.bounds = (-0.001,1000)
continue
# Close uptake rates for the rest of the boundaries
if rxn.id.startswith("EX_"):
rxn.bounds = (0,1000)
if rxn.id.startswith("SK_"):
rxn.bounds = (0,1000)
if rxn.id.startswith("DM_"):
rxn.bounds = (0,1000)
##### ----- Save iCHO3K unblocked version ----- #####
# XML
model_name_xml = f'../iCHO3K/Model/iCHO3K_{suffix}_unblocked.xml'
write_sbml_model(model_unblocked, model_name_xml)
# JSON, because the sbml doesnt save the subsystems
model_name_json = f'../iCHO3K/Model/iCHO3K_{suffix}_unblocked.json'
save_json_model(model_unblocked, model_name_json)
# MATLAB
model_name_matlab = f'../iCHO3K/Model/iCHO3K_{suffix}_unblocked.mat'
save_matlab_model(model_unblocked, model_name_matlab)Many figures in Analyses/ and the accompanying manuscript (https://doi.org/10.1101/2025.04.10.647063) were generated from notebooks in Python/:
- Reconstruction comparisons and Gene Essentiality Analysis →
Python/Comparison of Metabolic Reconstructions.ipynb→ Analyses/recons_comparisons/ - Subsystem coverage & sunbursts →
Python/Comparison of Computational Tests.ipynb /1. Subsystem Overview and Analysis→ Analyses/subsystem_overview/ - Topology & t-SNE →
Python/Comparison of Computational Tests.ipynb /4. tSNE Comparison of Models→ Analyses/tSNE/ - Growth rate prediction (WT vs ZeLa) →
Python/Comparison of Computational Tests.ipynb /5. pFBA→ Analyses/growth_rate_pred/
Most notebooks begin with a “Paths & Environment” cell—update paths as needed. For strict reproducibility, pin exact package versions via
environment.ymland use releases/DOI snapshots.
Curated inputs and derived artifacts are organized under Data/. Key elements:
- WT & ZeLa context-specific models →
Data/Context_specific_models/ - Experimentally validated CHO essential genes →
Data/Gene_Essentiality/ - GPR from Recon3D to CHO mapping →
Data/GPR_Mapping/ - Metabolite Curation Files →
Data/Metabolites/ - Soruce Reconstructions Datasets and Models →
Data/Metabolites/ - ZeLa vs WT Fed-batch Process Raw Data →
Data/ZeLa Data/ - ZeLa vs WT Fed-batch Process pre-processed Data →
Data/Uptake_Secretion_Rates/
The models in
iCHO3K/Model/are the unblocked iCHO3K variants (blocked reactions removed; boundaries configured for context-specific extraction such as mCADRE).
To generate the full iCHO3K model directly from the Excel dataset, see Create iCHO3K Model.
- Excel: Final iCHO3K set of reactions, genes and metabolites lives in
iCHO3K/Datasetfor inspection and conversion. - COBRApy (JSON / SBML)
import cobra # JSON m_json = cobra.io.load_json_model("iCHO3K/Model/<model_name>.json") # SBML m_sbml = cobra.io.read_sbml_model("iCHO3K/Model/<model_name>.xml")
- LP/QP solvers: GLPK (free), CPLEX/Gurobi (commercial, academic licenses). Set via COBRApy:
import cobra cobra.Configuration().solver = "glpk" # or "gurobi", "cplex"
- Speed: Prefer commercial solvers for large sampling tasks; reduce model size using context-specific models; cache solutions where possible.
- Numerics: Tighten feasibility/optimality tolerances for sensitive analyses.
Contributions are welcome!
- Issues: Report bugs, request features, or flag data discrepancies.
- PRs: Use feature branches; include a clear description, minimal reproducible example or notebook, and updated docs.
- Style: PEP 8 for Python; strip heavy notebook outputs before committing.
- Data: Avoid committing large binaries; use Git LFS or attach to Releases/Zenodo.
If contributing new datasets or model variants, please include:
- Data dictionary (column descriptions, units)
- Provenance (source links/versions)
- Minimal script/notebook to regenerate derived artifacts
If you use iCHO3K or materials from this repository, please cite the bioRxiv preprint:
Di Giusto, P., Choi, D.-H., et al. (2025). A community-consensus reconstruction of Chinese Hamster metabolism enables structural systems biology analyses to decipher metabolic rewiring in lactate-free CHO cells. bioRxiv. https://doi.org/10.1101/2025.04.10.647063 (v1 posted April 17, 2025).
- Pablo Di Giusto — pdigiusto@health.ucsd.edu · pablodigiusto91@gmail.com
Systems Biology & Cell Engineering Lab (Lewis Lab), University of Georgia
We thank the iCHO3K community contributors and collaborators. This work builds upon public resources: iCHO1766, iCHO2101, iCHO2291, iCHO2441, Recon3D, Human-GEM, BiGG, MetaNetX, Rhea, UniProt, BRENDA, and others referenced throughout the notebooks and the manuscript.
I see .gitattributes LFS pointers instead of files.
Run:
git lfs install
git lfs pullSolver not found / poor performance.
Install an LP solver (GLPK works; Gurobi/CPLEX recommended for speed). Set cobra.Configuration().solver = "gurobi" once installed.
Model won’t optimize (infeasible).
- Harmonize exchange bounds across conditions.
- Check blocked reactions / dead-end metabolites.
- For comparative runs (WT vs ZeLa), ensure identical media constraints.
Notebook paths are wrong.
Edit the first “Paths & Environment” cell—most notebooks expose a single place to set root paths.