Metadata-Version: 2.5
Name: altar-legnet
Version: 0.2.0
Summary: LegNet model binding for large-scale Altar MPRA 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<0.3,>=0.2.0
Provides-Extra: test
Requires-Dist: pytest-asyncio>=0.24; extra == 'test'
Requires-Dist: pytest>=8; extra == 'test'
Description-Content-Type: text/markdown

# Altar LegNet binding

`altar-legnet` is the lightweight control-plane binding for large-scale human LegNet MPRA activity
variant scoring. It produces backend-neutral container plans and does not import PyTorch.

The binding scores hg38 SNVs in independently retryable Altar variant batches. Each variant contributes a
200 bp genomic context with the SNV at index 100; the runtime adds the published 5′ and 3′ 15 bp LentiMPRA
adapters before the 230 bp model inference. The stable scalar effect is signed `alt_activity - ref_activity`.

Checkpoint bytes and the matching training configuration are independent, content-addressed resources.
This supports the upstream K562, HepG2, and WTC11 models as well as compatible checkpoints without tying
Altar to Zenodo, Hugging Face, Chorus, or another provider. The isolated executor is in `runtimes/legnet`.
`LegNetConfiguration` uses the digest-pinned runtime tested with the binding by default; callers can override
it with another immutable image reference.
