Protein Structure Evidence Resolver

Resolve a public protein identifier to source-linked experimental PDB structures and existing AlphaFold DB models while keeping the two evidence classes separate.

What this does not do

• No claim that one structure is universally best.
• No docking or drug recommendation.
• No clinical or diagnostic interpretation.
• No structure-file redistribution beyond source terms.
• No predicted/experimental conflation.
• No private sequence upload in v1.
• No structure prediction; the Actor only retrieves existing official records.

Sources and authorization

The Actor makes bounded GET requests only to official RCSB PDB Search/Data and file services, official PDBe entry/mapping services, AlphaFold DB, and UniProt REST. It accepts identifiers, never user URLs. Metadata carries request locators, response hashes, and contract versions. Coordinate copying is opt-in and limited to official RCSB or AlphaFold DB mmCIF URLs under their published terms.

Input

Each of 1-500 targets contains exactly one uniprotId, pdbId, or alphaFoldId. Experimental evidence, predicted evidence, ligand metadata, and coordinates can be independently controlled. selectionCriteria applies transparent filters and lexical ordering only.

{
  "targets": [{"uniprotId": "P69905"}],
  "includeExperimental": true,
  "includePredicted": true,
  "includeLigands": true,
  "includeCoordinates": false,
  "concurrency": 1
}

Output

One result row per processed target contains normalized mapping evidence, the exact UniProt sequence and SHA-256 when mapped, separate experimentalStructures and predictedModels, chains, assemblies, SIFTS-based sequence coverage and gaps, engineered/mutation indicators, observed ligands, experimental quality, AlphaFold confidence, conflicts, comparability, optional explicit selection, and source locators. A separate uncharged run-summary row reports counts and billing evidence.

Example shape:

{
  "entityKey": "P69905",
  "status": "resolved",
  "experimentalStructures": [{"evidenceClass": "experimental", "pdbId": "4HHB"}],
  "predictedModels": [{"evidenceClass": "predicted", "alphaFoldId": "AF-P69905-F1"}],
  "comparability": "comparable",
  "selection": {"applied": false, "orderedIdentifiers": [], "statement": "No ranking or best-structure claim was applied."},
  "charge": {"event": "structure-evidence", "chargeable": true}
}

Pricing

The exact primary event is structure-evidence at $0.03 per target. A 10-target run can charge at most 10 events, or $0.30, plus Apify platform usage. Resolved evidence, useful terminal no-evidence, ambiguity/conflict, and metadata-complete coordinate-partial results are chargeable. Invalid IDs, mapping invariant failures, rights blocks, source outages without a terminal result, internal errors, and cap stops are not.

State and baseline behavior

The Actor is stateless. It does not read or commit baselines and does not mutate source records.

Partial failures and negative results

Source outages are visible and uncharged when no terminal result can be established. A metadata result can remain chargeable when optional coordinates fail, with partial_coordinates and an explicit warning. Official terminal absence is a useful no_evidence result. Conflicting mappings are returned as evidence, never silently resolved.

Resource and artifact limits

Launch memory is 256 MB minimum, 1024 MB default, and 2048 MB maximum; hosted timeout is 600 seconds. Concurrency is 1-8. JSON responses are capped at 4 MB and depth 40. At most 50 experimental structures are considered per target. Coordinates are capped at 3 artifacts per target, 5,000,000 bytes each, and 10,000,000 bytes total. KVS keys are generated, and every coordinate has content type, byte length, exact SHA-256, schema version, generation rule, and source URL in ARTIFACT_MANIFEST.json.

Data handling, privacy, and deletion

Inputs contain public identifiers only. No credentials or private sequences are accepted. Results and requested coordinate artifacts remain in the caller's Apify storage and follow that storage's retention/deletion controls. See PRIVACY.md.

Source rights and attribution

RCSB PDB and PDBe describe experimental PDB archive records; AlphaFold DB provides pre-computed predictions and confidence data; UniProt provides the reference accession and sequence. Users remain responsible for source terms and downstream redistribution. See SOURCE_POLICY.md and the provider contracts under source_contracts/.

Support

See SUPPORT.md. Candidate upload and Store publication are intentionally outside this implementation.