Protein Structure Release Monitor

Monitor named protein targets for added, revised, obsoleted, replaced, or confirmed-removed experimental structures and existing predicted models.

What this does not do

• No target discovery.
• No interpretation of biological significance.
• No single-check removal declaration.
• No docking or medical claims.
• No coordinate download by default.
• No independent source crawling.
• No structure prediction. AlphaFold DB rows are existing provider records and are always labeled predicted; PDB rows are always labeled experimental.

Official sources

The Actor makes bounded GET requests only to official RCSB PDB Search/Data API, PDBe API v2, AlphaFold DB API, and UniProt REST entry endpoints. It does not accept user URLs or credentials. SOURCE_POLICY.md records the exact endpoints, authority, attribution, and failure semantics.

RCSB and PDBe catalogs are reconciled before experimental additions become material changes. An ID visible in only one catalog is reported under catalogChurn and is not treated as an addition or removal.

Input

Each target contains exactly one uniprotId, pdbId, or alphaFoldId. A persistent run also requires a user-selected baselineStoreId.

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

removalConfirmationHours defaults to and cannot be lower than 6. concurrency defaults to 4 and cannot exceed 8. A run accepts 1–500 unique targets. At least one of experimental or predicted evidence must be enabled.

Output

The default dataset receives one bounded result row per completed target and one uncharged run-summary row. A typical result contains:

{
  "recordType": "result",
  "schemaVersion": "1.0.0",
  "status": "changed",
  "finding": "material structure evidence changed",
  "charge": {"event": "target-check", "priceUsd": 0.03, "chargeable": true},
  "evidenceSummary": {
    "uniprotAccession": "P69905",
    "experimentalCount": 12,
    "predictedCount": 1,
    "obsoleteOrReplacedCount": 0
  },
  "changes": [{"changeClass": "ligand_set_changed", "recordId": "4HHB"}]
}

Change records use stable classes: new_structure, metadata_revised, sequence_mapping_changed, ligand_set_changed, confidence_changed, model_version_changed, obsolete, replaced, missing_unconfirmed, and removed_confirmed. Every change includes exact normalized field paths and official-source locators. Full snapshots and detailed diffs remain in the named baseline store or KVS artifact rather than a multi-megabyte dataset cell. ARTIFACT_MANIFEST.json records content type, byte length, SHA-256, schema version, and generation rule.

Pricing

The primary pay-per-event event is exactly target-check at $0.03 per target check. Maximum event cost is number of targets × $0.03; for 100 targets the maximum is $3.00, subject to the platform charge cap.

Baseline creation, unchanged checks, material changes, official obsolete/replaced states, confirmed removals, and completed no-evidence checks are chargeable. Invalid targets, mapping failures, source outages, rate-limit exhaustion, parse failures, and first unconfirmed missing observations are not charged. The synthetic apify-default-dataset-item event must remain disabled.

Baseline and removal behavior

Persistent mode uses the caller-selected KVS. For chargeable results the Actor writes a PREPARED journal, pushes exactly one target-check, verifies charged_count == 1, then writes immutable history and swaps the canonical baseline. A zero charge, outage, failure, or charge cap never advances the baseline.

Absence is not removal. The first successful missing observation is uncharged, updates only a separate missing journal, and leaves the canonical baseline intact. The same missing-record set must be observed by successful provider checks at least six hours later before removed_confirmed. Restoration clears the journal after a charged successful check. An official PDBe obsolete/replacement state is evidence and does not require the two-observation delay.

Partial failures and negative results

An HTTP 404/410 from the exact entry endpoint is a terminal source fact only where the contract permits it. Timeouts, 429 responses, malformed 200 responses, response limits, redirects outside official hosts, and provider disagreements are never converted into missing evidence. Any requested provider outage makes the target row uncharged and leaves all monitor state unchanged.

Resources

Candidate resources are minimum 256 MB, default 1024 MB, maximum 2048 MB, with a 600-second hosted timeout. Source JSON is capped at 8 MB and depth 40. RCSB catalogs are bounded at 1,000 identifiers; if truncated, removals are disabled and prior records are carried forward.

Data handling, privacy, and deletion

Inputs contain public protein identifiers and a caller-selected Apify KVS reference. No credentials or user URLs are accepted. Raw provider bodies are hashed in memory and are not retained. Dataset rows contain normalized public metadata and redacted input. Delete the run dataset/default KVS and the selected baseline KVS through Apify to remove stored results and history. See PRIVACY.md.

Source rights and attribution

Results preserve provider names and evidence URLs. Source data remains subject to RCSB PDB, PDBe/EMBL-EBI, AlphaFold DB, and UniProt terms and attribution requirements. The Actor does not transfer ownership or certify downstream rights. See SOURCE_POLICY.md.

Limitations

Catalog releases are asynchronous, provider fields can drift, and a catalog cap can delay change recognition. Confidence changes are provider-reported metadata changes, not biological interpretation. Coverage is a deterministic residue-range summary against the exact UniProt sequence identity; no structural alignment or quality ranking is performed.

Support and changelog

See SUPPORT.md for support and CHANGELOG.md for version history.