Metadata-Version: 2.5
Name: altar-chrombpnet
Version: 0.8.1
Summary: ChromBPNet model binding for large-scale Altar variant scoring
Project-URL: Documentation, https://kundajelab.github.io/altar/
Author: Riya Sinha
License-Expression: MIT
License-File: LICENSE
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Topic :: Scientific/Engineering :: Bio-Informatics
Classifier: Typing :: Typed
Requires-Python: >=3.12
Requires-Dist: altar-identity<0.9,>=0.8.1
Requires-Dist: altar<0.9,>=0.8.1
Provides-Extra: inspect
Requires-Dist: h5py>=3.10; extra == 'inspect'
Provides-Extra: test
Requires-Dist: h5py>=3.10; extra == 'test'
Requires-Dist: pytest-asyncio>=0.24; extra == 'test'
Requires-Dist: pytest>=8; extra == 'test'
Description-Content-Type: text/markdown

# altar-chrombpnet

Adds [ChromBPNet](https://github.com/kundajelab/chrombpnet) to Altar: predicted effects of variants on chromatin
accessibility in one cell type, from a five-fold model ensemble, run in a container on an NVIDIA GPU.

Validation: Reference comparison (compared with kundajelab/varscore). Interpretation: Provisional.

```bash
pip install altar-chrombpnet
```

Fetch ENCODE's K562 DNase ensemble and write its resource manifest, then derive the files scoring reads:

```bash
altar-chrombpnet-resources fetch-official --output-dir chrombpnet-resources
altar prepare spec.toml --model k562-dnase
```

A run spec section:

```toml
[models.k562-dnase]
type = "CHROMBPNET"
resource_manifest = "chrombpnet-resources/chrombpnet.resources.json"
```

ChromBPNet scores, prepares and interprets bias-corrected models only. The five fold files must be
`model.chrombpnet_nobias.fold_<k>.<experiment>.h5`, not the full `model.chrombpnet.fold_<k>.<experiment>.h5`, whose
predictions keep the Tn5 or DNase bias, and not a standalone bias model. Check them before you register a model:

```bash
pip install 'altar-chrombpnet[inspect]'
altar-chrombpnet-resources inspect fold_*/model.chrombpnet_nobias.fold_*.h5
```

It prints one JSON line per file, with its kind (`absent`, `included` or `bias_only`) and input and output
lengths, and exits 1 if any file is not bias-corrected (`absent`). Both scoring images a run can select refuse
such a fold file before loading it, too.

`altar models chrombpnet` lists the columns and configuration fields.

Documentation: https://kundajelab.github.io/altar/integrations/chrombpnet/

## Cite and terms

- ChromBPNet: Pampari A, et al. bioRxiv (2024), [doi:10.1101/2024.12.25.630221](https://doi.org/10.1101/2024.12.25.630221).
  The ChromBPNet code is MIT-licensed.
- The official K562 models are ENCODE data, free to use; [cite ENCODE](https://www.encodeproject.org/help/citing-encode/)
  and the accessions you used.

## Licence

MIT. See [LICENSE](https://github.com/kundajelab/altar/blob/main/bindings/chrombpnet/LICENSE).
