altar-opentargets-e2g

This binding normalizes the Open Targets Platform enhancer_to_gene dataset (ENCODE-rE2G evidence) into Altar's public VariantGeneLinkSource relation.

The package owns Open Targets schema and provenance semantics, but not storage. Convert source rows to OpenTargetsE2GRow, stream them through OpenTargetsE2GImporter, and query the published generation with OpenTargetsE2GSource. Both operate on Altar's generic VariantGeneLinkStore; no Open Targets × PostgreSQL, Parquet, DuckDB, or service-specific class is required.

from altar.sources import InMemoryVariantGeneLinkStore
from altar_opentargets_e2g import (
    OpenTargetsE2GImportConfig,
    OpenTargetsE2GImporter,
    OpenTargetsE2GSource,
)

store = InMemoryVariantGeneLinkStore()
await OpenTargetsE2GImporter(store).ingest(
    rows,
    OpenTargetsE2GImportConfig(
        data_release="25.12",
        score_threshold=0.6,
        threshold_applied=True,
    ),
    generation_id="sha256-of-the-source-asset",
)
source = OpenTargetsE2GSource(store)

The importer accepts an iterable so operators can read large upstream Parquet datasets in bounded batches. OpenTargetsE2GRow.from_source_mapping() accepts the official camelCase enhancer_to_gene Parquet fields; the binding intentionally does not require a particular Parquet engine. See the Open Targets dataset documentation and rE2G documentation for acquisition and upstream transformations.

The Open Targets pipeline currently filters released rE2G output at score >= 0.6. Pass the threshold and whether it was applied in OpenTargetsE2GImportConfig; Altar retains those semantics in every LinkScore rather than treating the released table as unthresholded raw predictions.

The caller supplies a generation identifier for the exact extraction; partial generations stay hidden, exact replay is idempotent, and publication happens only after the iterable completes. Use VariantGeneLinkQuery.data_releases when a store contains more than one Open Targets release.

The contract is imported from altar; the binding is independently versioned and tested against the public Altar conformance suite.