Skip to main content

Altar Sei binding

altar-sei is the lightweight control-plane binding for high-throughput Sei variant scoring. It registers SEI in Altar's canonical model-plugin group and does not install PyTorch, SciPy, the model architecture, or the official resources.

Each model instance identifies the official checkpoint, projection vectors, histone-profile indices, sequence-class metadata, and an immutable runtime image by SHA-256 digest. The image defaults to altar_sei.DEFAULT_IMAGE, a published image validated against FunctionLab's official Sei framework (runtimes/sei/validation/cpu-official-81f595f.json). The binding emits one independently retryable task for each staged batch, and the runtime loads Sei once and microbatches every reference/alternate sequence pair. prepare_scoring_batches cache-filters Sei batches when it is given the DetailStore that holds sequence_class_effects.

The primary row reports the class with the largest absolute effect while retaining its sign. The sequence_class_effects detail table preserves all 40 published class effects. Effects reproduce the official workflow: strand-averaged 21,907-profile predictions are nucleosome-occupancy adjusted as a reference/alternate pair, projected onto the published class vectors, and subtracted as alternate minus reference.

The per-profile predictions behind those effects are an opt-in artifact. A run whose output selection is SeiPlugin.profile_effects_output also writes chromatin_profile_effects: one .npz per batch (application/x-npz) with variant_ids, the float32 effects matrix of strand-averaged alternate-minus-reference predictions for all 21,907 profiles (before nucleosome adjustment, NaN for an unscorable variant), the profile_index of each column, and profile_names, the published name of each column's profile, so a consumer maps columns to profiles without the runtime or its resources. The names come from the official archive's target.names, which SeiConfiguration.target_names identifies by digest; scores never read it, so only a run that selects the artifact requires it and stages it. It is not part of the score lineage. Altar never decodes the archive; the scoring engine digests the file and records it in an ArtifactStore with the variants it covers, and batch preparation treats a variant without it as a cache miss. The default selection's command line is unchanged; selecting the artifact is a different output selection and therefore a different score lineage. In an altar run spec, name the artifact in the model's output and declare an [artifact_store].

The initial runtime accepts SNVs on hg19 or hg38. The manifest declares SNV-only variant eligibility, so Altar's batch preparation filters out and reports other variant classes before they reach the runtime. The runtime records incomplete windows, reference mismatches, unsupported allele symbols, and unsupported IUPAC window bases as null-score rows that keep the variant's identity, logs the reason, and continues to score valid batch neighbors; the primary table holds scores only. Such a row has n_classes_scored 0 and no class rows. The binding declares n_classes_scored as the row count of sequence_class_effects, so batch preparation counts the variant as complete on later runs. It also writes a null-score row for an indel in a batch staged without that filter, but Altar's scoring engine rejects that row and fails the batch rather than store a score for a variant the manifest excludes. A repeated canonical variant fails the batch, as in every other runtime. N bases in otherwise valid windows use Selene's uniform encoding and are flagged in contains_unk. That is narrower than Selene's general indel machinery and is explicit: unsupported alleles do not receive an unversioned alignment policy. The operational max_abs_class_effect >= 1 prioritization default is not a universal biological cutoff; consumers should calibrate prioritization for their cohort and question.

The executor lives under runtimes/sei. The official Sei implementation and resources are restricted to academic and research use; those terms apply independently of this MIT-licensed binding.

Metadata

Release files for altar-sei 0.3.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for altar-sei 0.3.0
File Size Uploaded Provenance Actions
altar_sei-0.3.0.tar.gz 19.3 kB Without provenance Details for altar_sei-0.3.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for altar-sei 0.3.0
File Interpreter ABI Platform Actions
altar_sei-0.3.0-py3-none-any.whl Python 3 none any Details for altar_sei-0.3.0-py3-none-any.whl

Total release size: 28.6 kB

Release files / altar_sei-0.3.0.tar.gz

Download URL altar_sei-0.3.0.tar.gz
Size 19.3 kB
Tags Source
SHA-256 checksum
How to use checksums
5eb0aaae82add0e5d756bfe349bc7986982c7478183d819f3d5594f0fcee0bee
BLAKE2b-256 checksum
How to use checksums
21a94e5bdc63e3664ac3bd085838b06b831ef63e941ed76f532d9c47a6638d3f
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via uv/0.12.17 {"installer":{"name":"uv","version":"0.12.17","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"22.04","id":"jammy","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

Release files / altar_sei-0.3.0-py3-none-any.whl

Download URL altar_sei-0.3.0-py3-none-any.whl
Size 9.3 kB
Tags Python 3
SHA-256 checksum
How to use checksums
43a15ab9f0a6315f9153aee0697c55b255537cfd90ab2c53918dbb651c63fe48
BLAKE2b-256 checksum
How to use checksums
7c30a1f017312d6c16d08a8fc14e38fe830b783d7d12e8084d8f869ca2e997bc
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via uv/0.12.17 {"installer":{"name":"uv","version":"0.12.17","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"22.04","id":"jammy","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

Security for altar-sei 0.3.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page