reactome-sim

This repository builds a simple workflow for SBML pathway simulation and parameter fitting with libRoadRunner.

Project Workflow

1. Start from an SBML model in working_homo-sapiens/.
2. Generate target values for all model species with generate_target_file.py via local Ollama.
3. Run optimization with roadrunner_new.py to fit selected model parameters against target values.
4. (Optional) Save optimized parameter values back into the SBML file.

Repository Structure

• working_examples/: Generic SBML templates for testing and experimentation.
• working_homo-sapiens/: Pathway SBML models prepared for dynamic simulation.
• generated_target/: Auto-generated prompts and CSV targets (created by generate_target_file.py).
• generate_target_file.py: Parses SBML, builds a prompt, calls local Ollama, and writes a target CSV.
• roadrunner_new.py: Runs OpenAI-ES style optimization with RoadRunner against target values.
• test.csv: Example target file used by roadrunner_new.py in its current configuration.

Requirements

• Python 3.10+
• Dependencies from requirements.txt
• Ollama installed locally
• A pulled Ollama model (default in script: llama3.2:3b)

Install dependencies:

pip install -r requirements.txt

Install and Run Ollama (Windows)

1) Install Ollama

Preferred (winget):

winget install --id Ollama.Ollama -e --accept-package-agreements --accept-source-agreements

If winget is unavailable, install from: https://ollama.com/download/windows

2) Open a new PowerShell and verify CLI

ollama --version

If you get ollama non riconosciuto, use the full path:

& "$env:LOCALAPPDATA\Programs\Ollama\ollama.exe" --version

3) Start the Ollama server

ollama serve

If ollama is not in PATH, use:

& "$env:LOCALAPPDATA\Programs\Ollama\ollama.exe" serve

Keep this terminal open while generating targets.

If you see bind ... 11434, it means the server is already running (this is okay).

4) Pull the model

In a second terminal:

ollama pull llama3.2:3b

PATH fallback:

& "$env:LOCALAPPDATA\Programs\Ollama\ollama.exe" pull llama3.2:3b

5) Verify server and models

Invoke-RestMethod http://127.0.0.1:11434/api/tags

You should see your downloaded model in the JSON response.

How To Run generate_target_file.py

1) Ensure Ollama server is running and model is pulled

Server:

ollama serve

Model:

ollama pull llama3.2:3b

2) Generate prompt + targets from SBML

python generate_target_file.py --sbml working_homo-sapiens/R-HSA-1855192.sbml

Default outputs:

• generated_target/R-HSA-1855192/prompt_R-HSA-1855192.txt
• generated_target/R-HSA-1855192/target.csv

Useful options

• Choose a specific Ollama model:

python generate_target_file.py --sbml working_homo-sapiens/R-HSA-1855192.sbml --model llama3.2:3b

• Dry run (no API call, prompt only):

python generate_target_file.py --sbml working_homo-sapiens/R-HSA-1855192.sbml --dry-run

• Custom output CSV:

python generate_target_file.py --sbml working_homo-sapiens/R-HSA-1855192.sbml --output-csv test.csv

How To Run roadrunner_new.py

roadrunner_new.py currently uses:

• SBML model: working_homo-sapiens/R-HSA-1855192.sbml
• Targets file: ./test.csv
• Tuned parameters: log_K_in, log_K_out, log_lambda_1

Run:

python roadrunner_new.py

What it does:

1. Loads target values from test.csv.
2. Simulates the model with RoadRunner.
3. Optimizes parameters with an OpenAI-ES loop.
4. Plots optimization loss.
5. Writes optimized parameter values back into the SBML file.

Practical End-to-End Example

1. Generate a target file:

python generate_target_file.py --sbml working_homo-sapiens/R-HSA-1855192.sbml --output-csv test.csv

2. Fit model parameters against that target file:

python roadrunner_new.py

How To Run merge_sbml.py

Use merge_sbml.py to merge two SBML pathway files into a single deterministic output model.

Basic usage

python merge_sbml.py working_homo-sapiens/R-HSA-1660537.sbml working_homo-sapiens/R-HSA-1660508.sbml -o working_homo-sapiens/merged_1660537_1660508.sbml

What gets written

• Merged SBML file at the path passed to -o.
• Merge report JSON next to it with suffix .merge_report.json.

For the example above, outputs are:

• working_homo-sapiens/merged_1660537_1660508.sbml
• working_homo-sapiens/merged_1660537_1660508_merge_report.json

Useful option

• Suppress validation warnings:

python merge_sbml.py working_homo-sapiens/R-HSA-1660537.sbml working_homo-sapiens/R-HSA-1660508.sbml -o working_homo-sapiens/merged_1660537_1660508.sbml --suppress-validation-warnings