No-code PhosKinTime dashboard

The unified dashboard is a Streamlit application for browsing completed PhosKinTime runs and for launching existing CLI/Pixi workflows without editing Python code. It is intentionally a thin user interface: scientific fitting, ODE solving, optimization, and analysis remain in the existing backend modules and scripts.

Launching the dashboard

From the repository root, start the dashboard with Pixi:

pixi run dashboard

For development with Streamlit reload-on-save enabled:

pixi run dashboard-dev

The tasks are defined in pixi.toml and run streamlit run dashboard/app.py with PYTHONPATH=. so the in-repository packages are importable.

Supported workflows

The launcher registry exposes only structured, known commands. It builds argument lists, not shell strings, and runs subprocesses from the repository root.

The dashboard does not invent new CLI flags. If an option is not listed in the workflow registry, it is not generated.

Result directory contract

Each dashboard-launched run uses a predictable result directory under results/<workflow>/<run_id>/ unless a workflow-specific backend chooses a compatible location. The browser expects this contract:

results/<run_id>/
├── metadata.json
├── command.txt
├── console.log
├── config_resolved.yaml
├── tables/
├── plots/
├── logs/
├── reports/
└── artifacts/

The result browser also recognizes practical legacy layouts, including Networkmodel top-level CSVs such as scalar_objective.csv, convergence_history.csv, prediction CSVs, optimization/, profiles/, posterior/, and legacy kinopt_results.xlsx or tfopt_results.xlsx files.

Upload behavior

The upload panel accepts .csv, .tsv, .xlsx, .yaml, .yml, .json, and .txt files for upload/preview or preset workflows. Files assigned to workflow --conf fields currently must be .toml because the target runners parse TOML. Uploaded files are copied to a per-run folder:

dashboard_uploads/<run_id>/

Uploads are not written into data/ and are not modified silently. The dashboard checks for empty files, unsupported extensions, unreadable files, and duplicate filenames after sanitization. CSV, TSV, and Excel inputs can be previewed with pandas before they are assigned to a workflow.

Workflow-specific validation follows the backend readers. ProtWise protein input and Networkmodel kinase network, TF network, MS/protein, RNA, and phosphoproteomics inputs must be CSV files because those runners read them with pandas.read_csv. Excel remains valid only for workflow inputs that are actually read as Excel, such as ProtWise phosphosite/RNA files or previous KinOpt/TFOpt result workbooks.

Command preview and execution

Before a run starts, the dashboard shows the exact command preview and the underlying argument list. This keeps runs reproducible and helps users copy the command into a terminal if preferred.

During execution, stdout and stderr are streamed into the page and written to console.log in the run directory. The runner records command.txt and launcher metadata in metadata.json. Failures are displayed with the return code and a log tail rather than crashing the Streamlit session.

Result browser usage

Use the Browse results tab to select an existing result directory. The browser displays:

• metadata.json with st.json when valid;
• command.txt, console.log, and other logs as text/code;
• CSV, TSV, and Excel tables with st.dataframe;
• PNG, JPG, JPEG, SVG, and HTML plots;
• Markdown, HTML, and PDF reports where Streamlit can render them;
• artifacts as downloadable files;
• a ZIP download button for the selected result directory.

If standard contract files are missing, the browser reports them clearly and still attempts to show recognized legacy outputs.

Workflow-specific panels

The result browser includes workflow tabs for KinOpt, TFOpt, ProtWise, Networkmodel, and advanced analyses.

• KinOpt displays discovered KinOpt result workbooks, alpha/beta-related tables where available, observed-vs-estimated output files, and existing plots.
• TFOpt displays TFOpt result workbooks, TF alpha/beta or equivalent tables, latent activity/dominance/knockout outputs when present, and existing plots.
• ProtWise displays ODE prediction tables, residual/fits files, sensitivity outputs, and generated plots or reports.
• Networkmodel supports both the standardized result layout and legacy Networkmodel outputs such as dashboard_bundle.pkl, scalar objective CSVs, convergence history, predictions, optimization summaries, profiles, and posterior outputs.
• Advanced analysis exposes parameterized command wrappers for existing scripts such as curve similarity, protein accumulator detection, subnetwork export, mechanistic insights, temporal sensitivity, and mechanism comparison. These analyses are not run automatically on page load.

Optional visualization packages are imported only when the affected panel is used. If a dependency is unavailable, the dashboard should show a clear message instead of failing during import.

Examples

Browse an existing result directory

1. Run pixi run dashboard.
2. Open Browse results.
3. Enter the base results folder, for example results.
4. Select a run folder.
5. Review metadata, tables, plots, logs, reports, artifacts, and workflow-specific tabs.

Run KinOpt

1. Open Run workflow.
2. Select KinOpt local (kinopt-local).
3. Upload or select phosphosite/protein and kinase-network inputs as required by the current CLI/config.
4. Preview uploaded tables.
5. Confirm validation passes.
6. Review the command preview.
7. Click Run workflow.
8. When the process completes, the result browser opens the run directory.

Run TFOpt

1. Select TFOpt local (tfopt-local).
2. Upload or select RNA/mRNA and TF-network inputs.
3. Pass a config file through the existing config argument when needed.
4. Review and run the generated command.
5. Inspect TFOpt tables, plots, logs, and workflow tabs after completion.

Run ProtWise/model

1. Select ProtWise model (protwise-model).
2. Assign the required protein/phosphosite inputs and optional config file.
3. Preview the command and run it.
4. Use the ProtWise panel to inspect predictions, residuals, fits, sensitivity files, plots, and reports.

Run Networkmodel

1. Select Networkmodel (networkmodel).
2. Assign kinase-network, TF-network, omics, and config inputs supported by the CLI.
3. Review the structured command preview.
4. Run the workflow.
5. Inspect scalar objective, convergence, predictions, optimization/profile/posterior outputs, plots, and the dashboard bundle when available.

Download a result ZIP

1. Open a result directory in Browse results.
2. Open the Download tab.
3. Click Prepare ZIP archive.
4. Click Download result ZIP.

The ZIP is generated in memory and is not committed to the repository.

Troubleshooting