InstaWell

Tools for organizing, processing, and visualizing thermal shift assay (TSA) data.

Why?

"Man I hate copying and pasting stuff in excel... I wish I could very quickly get from raw TSA data + layout to dose-response curves and Tm values without a million clicks."

Features

• Flexible layouts → parse arbitrary condition fields (e.g., concentration | ligand | protein | buffer)
• Long/wide transforms and replicate averaging
• Background subtraction using a non-protein control (NPC) marker
• Min-max scaling and derivative-based min-temperature (Tm-like) extraction
• Prism-style 4PL fit (log10 domain with logEC50) + diagnostics CSV
• Plotly figures: raw per-well, processed averages, and min-temperature scatter
• Jupyter widgets to browse generated figures interactively

Install

pip install instawell
# Optional: notebook extras (recommended)
pip install 'instawell[notebook]'

Python ≥3.10 recommended, as well as the tool uv for development environments.

Or install from source:

git clone https://github.com/DavidHein96/InstaWell.git
cd InstaWell
uv sync

Dash App

Prefer a UI? Install the Dash extra and launch the bundled server:

pip install 'instawell[dash]'
instawell-dash --port 8050 --experiments-root experiments

You can tweak --host, --port, --experiments-root, and --debug. The app exposes layout upload/validation, well filtering, and figure browsing directly in the browser. See docs/dash_app.md for architecture details and workflow tips.

What does this do?

Instawell takes your instrument's TSA output (temperatures × wells) and a layout that describes each well's condition (e.g., concentration|ligand|protein|buffer). It then organizes, QC-plots, and computes dose-response summaries.

1) Inputs

A) TSA output (raw instrument CSV) Each column after the temperature is a well (A1, A2, …). Values are the signal (e.g., fluorescence).

Temperature,A1,A2,A3,B1,B2,B3
25.0,  120,115,118,  130,126,127
26.0,  118,114,117,  128,125,126
...

B) Layout (maps well positions to conditions) This ties each well to a condition string your pipeline understands.

well_row,1,2,3
A, 0|DMSO|NPC|PBS,  1.5|DrugX|ProteinA|PBS,  6|DrugX|ProteinA|PBS
B, 0|DMSO|NPC|PBS,  1.5|DrugX|ProteinA|PBS,  6|DrugX|ProteinA|PBS

• well_row + column number → well name (e.g., A1, A2, …).
• Condition fields (order matters): concentration | ligand | protein | buffer
• NPC in protein = non-protein control for background.

---

2) What the pipeline does (high level)

1. Ingest & organize

   • Melts the wide TSA matrix into long format (one measurement per row).

   • Expands condition fields into columns and builds:

     • unqcond = concentration|ligand|protein|buffer
     • well_unqcond = well|concentration|ligand|protein|buffer

   • 📄 01_raw_organized_data.csv

   Example (long rows):

   Temperature,well,value,concentration,ligand,protein,buffer,unqcond,well_unqcond
   25.0,A2,115,1.5,DrugX,ProteinA,PBS,1.5|DrugX|ProteinA|PBS,A2|1.5|DrugX|ProteinA|PBS
   26.0,A2,114,1.5,DrugX,ProteinA,PBS,1.5|DrugX|ProteinA|PBS,A2|1.5|DrugX|ProteinA|PBS
   ...

2. (Optional) Filter bad wells

   • You inspect raw per-well plots to flag odd traces (bubbles, spikes, drifts).
   • 📊 raw_figures_widget(...) helps you browse quickly.
   • 📄 02_filtered_organized_data.csv (even if nothing removed, for traceability)

3. Average replicates

   • Averages value over wells sharing the same unqcond at each temperature.
   • Produces long + pivoted wide tables keyed by Temperature.
   • 📄 03_averaged_data.csv, 03_averaged_data_long.csv

4. Background subtraction (NPC)

   • For each (ligand, buffer, concentration) with protein ≠ NPC, subtracts the matching NPC column.
   • Leaves NPC columns out of the final set (less heavy-handed removal).
   • 📄 04_bg_subtracted_data.csv, 04_bg_subtracted_data_long.csv

5. Min–max scaling (QC convenience)

   • Scales each unqcond trace to [0, 1] to make shapes comparable.
   • 📄 05_min_max_scaled_data.csv, 05_min_max_scaled_data_long.csv

6. Derivative & min temperature (“Tm-like”)

   • Computes derivative curves and finds the temperature at the minimum derivative per unqcond.
   • 📄 06_derivative_data(_long).csv
   • 📄 07_min_temperatures.csv (has concentration, ligand, protein, buffer, min_temperature)

7. Dose–response (Prism-style 4PL) EXPERIMENTAL

   • Fits a 4-parameter logistic in log10 dose space using logEC50 (zeros are excluded there).
   • This is currently not well tested and is experimental—use with caution!
   • Outputs parameter table (Bottom, Top, logEC50, EC50, Hill, SEs, 95% CIs, RSS/RMSE, AIC/BIC) and point-wise diagnostics.
   • 📄 08_curve_params.csv, 08_curve_diagnostics.csv

Documentation

The full documentation now lives in the Quarto site under docs/. To browse locally:

quarto preview docs

Key pages:

• docs/index.md – overview & quick start
• docs/pipeline.md – numbered CSV pipeline guide
• docs/dash_app.md – Dash workflow, layout designer, and validation tips

Quick start

Also see the example notebook

from instawell import (
    setup_experiment,          # Step 00 - creates a folder structure and saves metadata
    ingest_data,               # Step 01 - organizes raw data and extracts conditions from the layout
    filter_wells,              # Step 02 - filtering of wells (required to be run)
    average_across_replicates, # Step 03 - groups replicate wells
    subtract_background,       # Step 04 - NPC background subtraction
    min_max_scale,             # Step 05 - min-max scaling
    calculate_derivative,      # Step 06 - derivative computation 
    find_min_temperature,      # Step 07 - min-temperature extraction
    calculate_curve_params,    # Step 08 - 4PL curve fitting
    load_experiment_context,   # load existing experiment context from disk
)

exp = setup_experiment(
    experiment_name="demo1",
    experiments_root="experiments/demo1",
    raw_data_path="data/demo1/raw.csv",
    layout_data_path="data/demo1/layout.csv",
    condition_fields=("concentration","ligand","protein","buffer"),
    condition_separator="|",
    empty_condition_placeholder="0",
    non_protein_control_marker="NPC",
)

# --- Pipeline (pass along the exp) ---
ingest_data(exp)                  # -> 01_raw_organized_data.csv
filter_wells(exp)                 # -> 02_filtered_organized_data.csv
average_across_replicates(exp)   # -> 03_averaged_data.csv (+ long for easier formatting)
subtract_background(exp)          # -> 04_bg_subtracted_data.csv (+ long)
min_max_scale(exp)                # -> 05_min_max_scaled_data.csv (+ long)
calculate_derivative(exp)         # -> 06_derivative_data.csv (+ long)
find_min_temperature(exp)         # -> 07_min_temperatures.csv
calculate_curve_params(exp)       # -> 08 curves: params/diagnostics CSVs (experimental)

Jupyter widgets (requires jupyter notebook)

from instawell import raw_figures_widget, processed_figures_widget, min_temp_figures_widget

# Raw per-well plots (discrete colors per well, only for raw and filtered data)
raw_figures_widget(exp)

# Processed/averaged plots (everything after averaging step)
processed_figures_widget(exp, data_source="bg_subtracted", color_scale="Thermal")

# Min-temperature scatter (with selectable modes in generator args experimental)
min_temp_figures_widget(exp, mode="log10_fit", color_scale="Viridis")

The widget wrappers preserve the docstrings & signatures of the underlying generators. Use show_help=True to display the docstring in a collapsible panel. You can also save individual figures from the generators by passing save_figs=True to the widget, since they are in plotly they are saved as HTML files by default, which you can open in a browser or convert to PNG using kaleido.

Typical files (by step)

experiments/<name>/

01_raw_organized_data.csv
02_filtered_organized_data.csv
03_averaged_data.csv
03_averaged_data_long.csv
04_bg_subtracted_data.csv
04_bg_subtracted_data_long.csv
05_min_max_scaled_data.csv
05_min_max_scaled_data_long.csv
06_derivative_data.csv
06_derivative_data_long.csv
07_min_temperatures.csv
08_curve_params.csv
08_curve_diagnostics.csv
experiment.log                  # A log file of pipeline steps
01_raw_plots/...                # saved HTML
03_averaged_plots/...
experiment_info.json            # Parsed layout and wells metadata
experiment.json                 # ExperimentContext configuration
filtered_wells.txt              # List of excluded wells for extra reference
original_raw_data.csv           # Copy of the original raw data
original_layout_data.csv        # Copy of the original layout data

Key concepts

• Condition fields - configured via ExperimentContext.condition_fields and joined with condition_separator to derive:

  • unqcond (unique condition string)
  • well_unqcond (unique well + condition)

• NPC background subtraction - subtracts a matching non-protein control column per panel; leaves NPC traces out of the final set.

Configuration (ExperimentContext)

Common fields:

• experiment_dir: output directory (CSV, plots, logs)
• raw_data_path, layout_data_path, temperature_column
• condition_fields: tuple of fields (order matters)
• condition_separator: string used to join/split fields (e.g., "|")
• empty_condition_placeholder: placeholder for 'blank' conditions. A fully blank well would be e.g., "0|0|0|0"
• non_protein_control_marker: e.g., "NPC"
• log_to_file, log_level: pipeline logging

Testing

# Run the full suite (unit, fuzz, Dash app, synthetic integration)
uv run pytest

# Run only Dash app tests
uv run pytest tests/test_dash_app.py tests/test_dash_integration.py

CI runs pytest --ignore=tests/test_integration_pipeline.py --ignore=tests/test_integration_tsa067.py because those two files depend on private TSA datasets (golden-file comparisons against real instrument data) that are not committed to the repository. All other tests run in CI on Python 3.10 and 3.12.

Development Notes

TODOs

• Expand the Dash app to cover more of the pipeline steps and improve the layout designer UX.
• Improve documentation coverage, especially around advanced usage and configuration options.
• Setup github pages for hosting the Quarto documentation site.
• Add more examples and tutorials in the docs.
• Optimize performance for larger datasets, especially in the data processing functions.

License

This project is licensed under the GNU AFFERO GENERAL PUBLIC LICENSE Version 3 (AGPL-3.0)