Dataset Viewer

The dataset viewer should be available soon. Please retry later.

AFDB v6 portable search shards

Current task, 30 September 2026: all 483 raw motif searches are complete. The main dataset remains motif-based protein pairs. Add optional AlphaFill geometric context, reusing existing structures and transplant metadata before new local/HPC computation. Do not group/filter by ligand identity, biochemical role, GO or activity. Missing context does not exclude baseline pairs.

Start with the runnable AlphaFill workflow, HPC handover, and allocation. The ge-01 inventory and accepted-endpoint overlap export are launched/queued. One real v4-to-v6 endpoint task extracted50 separate transplant alternatives; this is a parsing/registration check, not population coverage or accepted pairs. HPC owns missing-entry computation, optional reference recovery, geometry, pair compatibility and merge/upload. Publication does not imply HPC dispatch.

Keep natural protein endpoints and explicit pose provenance. Separate protein retention/identity from coordinate fixation; protected protein coordinates are still movable. No new H-bond/burial/hotspot, partner-chain, symmetry, unindexed motif or reaction modeling is included. Baseline validation remains incomplete and original-cluster owned until a coordinated cutover. All raw searches are reused. Readiness is explicit.

This repository is a progressively uploaded, independently sharded copy of the published afdb_uniprot_v6 Foldcomp collection and a GPBridge motif query bundle. It is intended to move uncapped catalytic-motif searches to CPU HPC systems.

The full source collection has 241,070,489 compressed structural records, divided into 483 shards, each containing at most 500,000 records. Coordinate payloads total 1,261,454,421,092 bytes before index/lookup packaging. This is the Foldcomp publication's coverage, not the complete nominal AFDB record count. Foldcomp excludes some problematic or discontinuous structures.

Uploads may be incomplete. manifest.jsonl describes the intended full source collection; only existing shards/*.tar files have uploaded. The initial priority is unfinished searches, so absence of a completed shard's coordinate TAR does not mean that shard was excluded from the research collection. completed_shards.json refers to searches finished on the original cluster, not to file-upload completeness or accepted training pairs.

Each TAR contains structures, structures.index, structures.lookup, structures.dbtype and manifest.json. Coordinate bytes are unchanged; offsets are adjusted within each shard. Structures over 1,024 residues remain only in the raw compressed input and are excluded from the active motif search index.

motifs_v2.tar.gz contains the existing 876-template M-CSA query library, PDB coordinates and annotations. The accepted two-site extension is a separate query campaign and is not included implicitly. Local search has no hit-count cap but does have explicit geometric/hash thresholds; this is not a claim of enumerating every possible physical conformation or proving catalytic function.

Raw retrieval hits still need current natural-sequence checks, complete ordered catalytic residue identity and joint all-atom geometry checks, and subsequent source–target alignment before forming GPBridge training pairs.

Code: https://github.com/ZhanghanNi/GPBridge

Sources and attribution

Upstream terms apply to their respective assets; no single new license is asserted over this mixed-source bundle. Motif geometry acceptance does not establish enzyme activity experimentally.

Downloads last month
6,580

Space using tonynzh2/afdb6 1