MPS2D Atlas — Membrane Protein Structure-to-Drug Atlas: pilot registry and benchmark release v0.9 (instance of record 2026-10-21)
收藏资源简介:
Pilot registry and benchmark release of the MPS2D Atlas, a programmatically curated, AI-ready database of every membrane protein with an experimental structure in the Protein Data Bank. Classifies all 80,873 UniProt accessions with a PDB structure by code from written, versioned rules; identifies 8,204 membrane proteins (7,883 integral, 321 associated) carrying 48,498 PDB entries and 70,352 entry–protein records. This release contains the registry (accession, classification, the rule that produced it, identifier-level evidence from PDBTM, OPM, mpstruc, MemProtMD, the RCSB membrane tag and UniProt annotation, and the PDB entries carried); the entry–protein table with deposited metadata; the cohort v3 benchmark set — 365 design-grade structures across ten families with the criterion values used for admission and stratification and structure-level TM-scores against four prediction arms spanning three architectures, in a declared transmembrane frame; the reconciliation-gate report (0 unaccounted entries against the five external sources); the instance ledger (every reported figure resolves to a ledger identifier); and the table schemas. Not in this release: the archive-wide computed metric layers (construct-fidelity indices SRF, RSF, DRF, EWI, SIF, fusion and partner indices beyond the cohort; per-residue resolved masks; TMbed topology; per-residue lDDT and distance arrays), which ship with Release 1. The pipeline source will be released under the MIT license at the public launch of the Atlas platform. Inputs fetched 2026-09-21; integrity gates (identifier canonicalization; experimental-only admission) adopted 2026-10-21; build commit fe2e665. Typed columnar tables with published schema and per-file SHA-256 checksums (CHECKSUMS.sha256, MANIFEST.json).



