Skip to content

Dashboard Usage

PhosKinTime includes an interactive Streamlit-based dashboard for visualizing the results of the global network model optimization.

Components

networkmodel/dashboard_app.py

The main Streamlit application. It loads result files from a specified output directory and renders:

  • Scalar objective diagnostics for the JAXopt/Diffrax run
  • Convergence history (fitness over generations)
  • Goodness of fit (predicted vs. observed for protein, RNA, phospho)
  • Residuals analysis
  • Image/video galleries of exported plots (PNG, JPEG, MP4)
  • PDF report viewer

networkmodel/dashboard_bundle.py

Saves and loads a compact binary bundle (dashboard_bundle.pkl) containing:

Field Description
args Command-line arguments from the runner
picked_index Index of the selected scalar-objective solution
frechet_scores Fréchet distance scores per protein
lambdas Lambda regularization values
solver_times ODE solver timing per protein
defaults Default parameter values
slices Parameter slices
xl / xu Lower / upper parameter bounds

The bundle stores scalar objective arrays and mode metadata, making it portable across Python sessions.

run_dashboard.py

Root-level launcher script. Sets up sys.path and calls networkmodel.dashboard_app.main().

Required Result Files

The dashboard looks for files in the --output-dir you specify. Required / optional:

File Status Description
dashboard_bundle.pkl Required Saved by networkmodel/runner.py after optimization
scalar_objective.csv Preferred Scalar objective values with mode metadata
pareto_F.csv Backward-compatible alias Scalar objective values
convergence_history.csv Optional Generation-by-generation convergence
pred_prot_picked.csv Optional Predicted protein time series (picked solution)
pred_rna_picked.csv Optional Predicted RNA time series (picked solution)
pred_phospho_picked.csv Optional Predicted phospho time series (picked solution)
*.png / *.jpg Optional Any image outputs from networkmodel/export.py
*.mp4 Optional Convergence animation
*.pdf Optional Report PDFs

If dashboard_bundle.pkl is missing the dashboard will fail with a FileNotFoundError.

How to Launch

streamlit run run_dashboard.py -- --output-dir results_model_global_distributive_knockout

Streamlit will start a local web server. The default URL is:

http://localhost:8501

Option 2: Via the networkmodel runner entry point

python -m networkmodel.runner runs networkmodel/runner.py. All settings default to config.toml values and can be overridden via CLI arguments:

# Run with defaults from config.toml
python -m networkmodel.runner

# Override specific settings
python -m networkmodel.runner \
  --output-dir results_global \
  --n-gen 500 \
  --solver jaxopt

Key arguments (all optional — defaults come from config.toml):

Argument Description
--kinase-net Path to kinase-substrate network
--tf-net Path to TF-gene network
--ms Path to MS protein data
--rna Path to RNA data
--output-dir Output directory
--n-gen Maximum JAXopt iterations
--solver jaxopt; legacy values are accepted and mapped with warnings
--sensitivity Enable sensitivity analysis
--scan Run hyperparameter scan

This runs the full optimization pipeline via networkmodel/runner.py and saves the bundle. After it finishes, launch the dashboard with Option 1 above.

Option 3: Python script

python run_dashboard.py --output-dir results_model_global_distributive_knockout

Common Failure Modes

Symptom Likely cause
FileNotFoundError: dashboard_bundle.pkl Run python -m networkmodel.runner first to generate results
KeyError on bundle fields Bundle was saved by an older version; re-run the optimizer
Streamlit not found Install streamlit: pip install streamlit
Empty plots Result CSVs are missing; check --output-dir path

Default Port

Streamlit defaults to port 8501. To use a different port:

streamlit run run_dashboard.py --server.port 8502 -- --output-dir results_model_global_distributive_jax

Relationship Between dashboard_app.py and dashboard_bundle.py

  • dashboard_bundle.py is a save/load utility — it knows nothing about Streamlit. It is called by networkmodel/runner.py at the end of optimization to persist results.
  • dashboard_app.py is the Streamlit UI — it calls dashboard_bundle.py's load_dashboard_bundle() at startup to restore the persisted results.

This separation ensures the dashboard can be launched independently from the optimization run, and makes the bundle serialization independent of UI concerns.

Inference Outputs

The dashboard includes an Inference tab when optional inference files are present. It displays:

  • optimization/best_fit.csv
  • optimization/multistart_summary.csv
  • optimization/multistart_parameters.csv
  • profiles/profile_likelihood_summary.csv
  • posterior/posterior_summary.csv
  • posterior/posterior_samples.csv
  • plots from plots/multistart/, plots/profile_likelihood/, and plots/posterior/

These are diagnostic scalar-objective summaries, not true multi-objective optimizer fronts.

Unified no-code dashboard

The newer unified dashboard lives in dashboard/app.py and can browse existing result directories, launch registered CLI/Pixi workflows, preview uploads, and show workflow-specific panels. See No-code PhosKinTime dashboard for launch commands, supported workflows, result-directory contract details, upload behavior, examples, and troubleshooting.