---
pretty_name: PhaseFlow-Peptide
license: other
tags:
  - peptide
  - liquid-liquid-phase-separation
  - phase-diagram
  - flow-matching
---

# PhaseFlow-Peptide

The `peptide/` package contains the short-peptide 4×4 PSSI phase-diagram data used
by the peptide branch of PhaseFlow. It contains the exact 280,778-row verified
PhaseFlow input table and the script needed to generate its missing-count
training views locally.

## Open access

All files in this package are openly available for research, benchmarking,
method development, and reproducible studies. We welcome community use and
redistribution; please cite the PhaseFlow work and retain relevant source
attribution.

## Data Structure

```text
peptide/
├── configs/
│   ├── dataset_contract.yaml
│   └── training.yaml            # best phase-prediction training setting
├── data/
│   ├── raw/
│   │   └── phase_diagram_original_scale.csv
├── metadata/
│   ├── data_dictionary.md
│   ├── dataset_summary.json
│   ├── feature_reproduction.md
│   └── file_inventory.csv
└── scripts/
    └── prepare_missing_splits.py
```

`data/raw/phase_diagram_original_scale.csv` is the direct PhaseFlow training
input. It has `AminoAcidSequence` plus `group_11` through `group_44`; NaN
denotes an unobserved PSSI-grid cell. Generate the required sibling
`by_missing/missing_<0-15>.csv` training views locally with the supplied script.

## Reproducibility

From this directory, generate the missing-count views locally:

```bash
python scripts/prepare_missing_splits.py
```

The final peptide PhaseFlow setting is in `configs/training.yaml`: Flow weight
32, language-model weight 0, batch size 2048, learning rate 0.0008, and
`missing_threshold` 15.

To train with the PhaseFlow peptide workflow, point `--data_path` at
`data/raw/phase_diagram_original_scale.csv`; its `by_missing/` sibling will
be discovered automatically.

## Provenance and limitations

The CSV is the earliest verified input available in this workspace. The
upstream, pre-CSV source and the historical conversion procedure were not
available, so this release does not assert a lineage beyond that table. It does
fully reproduce the missing-count views from the distributed CSV. The
package is for research and reproducibility, not clinical, diagnostic, or
experimental annotation use. Please cite the accompanying PhaseFlow manuscript
or public repository release when using or redistributing these data.