Breda-Charlton-FixationGrower

Code and data for the figures in:

Breda JR*, Charlton JA*, Willock JM, Kopec CD, Brody CD (2026).
FixGrower: An efficient and robust curriculum for shaping fixation behavior in rodents.
eLife — bioRxiv preprint

---

Setup

1. Create and activate the conda environment (Python 3.10)

conda env create -f environment.yml
conda activate fixation-grower

2. Register the Jupyter kernel

python -m ipykernel install --user --name fixation-grower --display-name "fixation-grower"

3. Data

Processed data files belong in data/ (not committed). Core fixation-curriculum files:

• trials_df.parquet, days_df.parquet, poke_df.parquet
• fig04_fixgrower_continued_stage11.parquet — Figure 4 stage-11 continued assessment
• fig05_cleaned_trials_df.parquet — Figure 5 DMS trial-level data
• fig05_fixgrow_days_to_target.parquet — Figure 5 per-animal days to fixation target
• fig06_mouse_fixation_trials.parquet — Figure 6 / Supplement S4 mouse trial-level data

See the figures folder for the associated publication figures.

---

Running a figure

Open any notebook in notebooks/ with Jupyter, select the fixation-grower kernel, and run all cells top-to-bottom. Figures are saved to figures/.

jupyter notebook notebooks/fig01.ipynb

---

Repository layout

Breda-Charlton-FixationGrower/
├── environment.yml                  # conda environment (Python 3.10)
├── pyproject.toml                   # installable package definition
├── data/                            # parquet data files (not committed)
├── figures/                         # figure outputs (not committed)
│   ├── figNN_FULL.png               # main-figure composite (reference; added manually)
│   ├── suppSN_FULL.png              # supplement composite (e.g. suppS3_FULL.png, suppS5_FULL.png)
│   └── figNN<letter>_….png          # panel PNGs reproduced by notebooks
├── notebooks/
│   ├── fig01.ipynb          # Figure 1: fixation growth example panels
│   ├── fig02.ipynb          # Figure 2: FixGrower grows faster to fixation target
│   ├── fig03.ipynb          # Figure 3: probe violation rates vs training time
│   ├── fig04.ipynb          # Figure 4: extended post-probe performance
│   ├── fig05.ipynb          # Figure 5: DMS task performance after FixGrower training
│   ├── fig06.ipynb          # Figure 6: FixGrower generalization to mice
│   ├── fig07.ipynb          # Figure 7: FixGrower vigor vs Legacy
│   ├── fig08.ipynb          # Figure 8: within-session engagement
│   ├── fig09.ipynb          # Figure 9: center-poke timing (fixation growth vs probe)
│   ├── suppS1.ipynb         # Supplement Figure S1: metrics by curricula phase + stage progressions
│   ├── suppS2.ipynb         # Supplement Figure S2: Legacy growth-ceiling simulations
│   ├── suppS3.ipynb         # Supplement Figure S3: raw fixation, violation, and rig data
│   ├── suppS4.ipynb         # Supplement Figure S4: extended mouse fixation data
│   └── suppS5.ipynb         # Supplement Figure S5: warm-up exclusion and full growth poke timing
└── src/fixation_grower/
    ├── config.py      # cohort definitions, colors, stage constants
    ├── paths.py       # data/ and figures/ directory resolution
    ├── io.py          # data loaders (load_trials_df, load_poke_df, …)
    ├── transforms.py  # feature engineering (days relative to stage, …)
    ├── plotting.py    # shared aesthetics and save_figure()
    ├── stats.py       # Legacy vs FixGrower statistical comparisons
    └── simulation.py  # Legacy growth-ceiling simulation (Supplement S2)

---

Notebook conventions

Each notebook follows a standard structure to maximize readability:

1. Figure overview (markdown) — # Figure N, embedded ![Full Figure](../figures/figNN_FULL.png), then ## Code For … with goal, panel table, and list of panel outputs the notebook generates
2. Imports cell — package imports only, no logic
3. Shared settings (optional) — constants reused across panels in the same notebook
4. Panel sections — one ## Panel X markdown header per panel, followed by its code
5. Cleared outputs — notebooks are committed without cell outputs

The figNN_FULL.png file is the full published figure for context; notebooks reproduce individual panels only.

Code placement rule: if logic is reused across figures or supplements it lives in src/fixation_grower/; if it is panel-specific constants and styling it stays inline in the notebook. No def statements in notebooks unless the function is fewer than 5 lines and used only once.

Terminology: curriculum arms are called Legacy and FixGrower throughout.

Data access: always use the package loaders (load_trials_df(), etc.) rather than bare pd.read_parquet() calls in notebooks.