Integrated Biochemical Reaction Network Generation and Pathway Analysis Toolkit
BioMathForge is a comprehensive toolkit for generating, integrating, and analyzing biochemical reaction networks. It combines AI-powered reaction network generation with advanced pathway analysis to provide insights into biological systems.
This toolkit is associated with the study: Tsutsui M et al., 2025. Literature-derived, context-aware gene regulatory networks improve biological predictions and mathematical modeling.
- AI-powered generation of reaction equations from biological inputs
- Automatic validation and integration of reactions into coherent networks
- Mass balance enforcement and divergence prevention for robust simulations
- Web-enhanced pathway inference using search-driven techniques
- Prediction and prioritization of expected readouts
- Graph-based modeling and analysis via LangGraph workflow
git clone https://github.com/okada-lab/BioMathForge
cd BioMathForge
pip install .
cp .env.example .env
# Edit `.env` and insert your valid API KEYsWe provide an example workflow using breast cancer-related equations curated from BioModels. These are located under examples/mcf-7.
Small-scale example equations (example_biomodels_equations.csv) derived from BioModels are provided. Full datasets and details are described in our accompanying paper and the repository okada-lab/context-dependent-GRN.
Convert raw equations to a standardized format using the generate_formatted_reactions function.
from dotenv import load_dotenv
load_dotenv()
from biomathforge import generate_formatted_reactions
import pandas as pd
biomodels_reactions = pd.read_csv("examples/mcf-7/example_biomodels_equations.csv")
validated_reactions = generate_formatted_reactions(biomodels_reactions)📄 Output: examples/mcf-7/generated_formula.txt
Use web-based research to identify key signaling pathways and expected readouts under experimental conditions.
from biomathforge import run_pathway_analysis
report = run_pathway_analysis(
reactions_path="examples/mcf-7/generated_formula.txt",
condition_path="examples/mcf-7/experimental_condition.txt"
)📄 Output: examples/mcf-7/pathway_analysis_result.json
Integrate the equations into a complete network using biological constraints and inferred readouts.
from biomathforge import integrate_reactions
equations = [eq.strip() for eq in open("examples/mcf-7/generated_formula.txt") if eq.strip()]
integrated_equations, source_nodes, sink_nodes = integrate_reactions(equations, report)📄 Outputs:
examples/mcf-7/integrated_equations_final.txtexamples/mcf-7/terminal_nodes.json
Add plausible feedback loops and crosstalk reactions to improve biological realism.
from biomathforge import run_enhance_feedback_crosstalk
report, enhancement_summary, added_reactions = run_enhance_feedback_crosstalk(
reactions_path="examples/mcf-7/integrated_equations_final.txt",
terminal_nodes_path="examples/mcf-7/terminal_nodes.json",
reactions_overviews_path="examples/mcf-7/pathway_analysis_result.json"
)
equations = report.split("\n")📄 Output: examples/mcf-7/breast_cancer_reactions.txt
Clean up, deduplicate, and finalize the set of reactions for downstream use (e.g., simulation or export).
from biomathforge import finalize_reactions
finalized_equations = finalize_reactions(equations)📄 Output: examples/mcf-7/breast_cancer_reactions_finalized.txt
- Python 3.10+
- OpenAI API key (or other supported LLM providers)
- TAVILY API key for web-enhanced pathway analysis (or other supported web search APIs)
This project is licensed under the MIT License.
If you use BioMathForge in your research, please cite our accompanying paper (coming soon) and reference this repository.