The evidence behind the Atlas.

Scientific scope

The versioned snapshot includes 30 human proteins (27 annotation-supported and three predicted PWWP-like candidates), nine selected nonhuman comparators, and 42 native annotation intervals. These are operational inventory counts, not experimental verification of every protein or domain. The source snapshot was assembled on 24 October 2026; version 1.2 adds scientific-source metadata and a production application candidate.

Coverage and classification

Protein association, domain sequence mapping, observed residues, complete backbone, complete side chains and biological ligand evidence are separate properties. There are 230 protein-associated PDB entries; 206 contain observed PWWP-core coordinates. Categories preserve the source's APO, Co-Crystals, Complexes and Unclassified Review labels. These categories organize entries and do not by themselves establish a ligand's biological role.

Source annotation cores may be shorter than folded domains or binding pockets. The 11OA example has nominally complete mapped core positions but lacks many standard heavy atoms, including aromatic side chains. Coordinate visualization does not repair unobserved atoms.

Structural comparisons

Gemmi 0.7.5 parses the first deposited model. Experimental-to-AlphaFold comparisons use fixed UniProt correspondence, positive-occupancy Cα atoms, and a proper-rotation Kabsch fit without outlier removal. Eligibility requires at least 40 common Cα atoms and 80% annotation-core coverage. The 205 eligible entry/domain representatives are an uneven, descriptive sample, not independent prediction benchmarks.

Cross-domain comparisons use US-align residue correspondence and both crop-length-normalized TM scores. The viewer applies an independently recomputed Kabsch least-squares transformation to the blue model. It does not use the different TM-optimized US-align transformation. All 153 human domain pairs remain available; the displayed comparison criterion is descriptive and does not establish pocket equivalence or selectivity. HDGF domain-swapped structures require assembly-aware interpretation.

AlphaFold and isoforms

131 source-assigned models are retained, including official canonical and isoform models and one separately labeled community model. Six isoform/domain corrections are documented. The local annotation-core pLDDT measures model confidence, not protein existence, biological function, or interdomain placement. The two primary computations reproduced from the minimal public inputs use 39 official canonical models; they do not repeat the full 131-model source and isoform audit.

Chemical identity and assay evidence

Full InChIKeys distinguish chemical identities. Preparation-level measurements remain separate from modeled stereochemical candidates, including compound 11. Direct affinities, displacement values, cellular responses, censored results and qualitative negatives retain distinct endpoints. The seven NSD3 counterscreens preserve four Kd >30 µM observations and three qualitative no-binding reports; qualitative negatives are not assigned numeric detection limits.

Contact evidence uses positive-occupancy heavy atoms within 5 Å in the first model and asymmetric unit; alternate conformers are retained and crystal symmetry is not expanded. The evidence graph's exact-contact projection does not include every assay. An exact identity/contact link does not transfer an assay result to the crystallographic construct.

Source-verification states

For reviewed headline records, the field-level overlay distinguishes verified, not reported in the inspected source scope, not inspected, not transcribed, not applicable and source conflict. A source conflict preserves differing statements, such as inconsistent replicate counts, without silently choosing one. Source access limitations remain visible. Other null assay fields remain uncurated. These annotations support interpretation and do not substitute for responsible scientific review.

Coordinates and molecular visualization

3Dmol.js 2.5.5 is served locally. Three frozen wwPDB coordinate examples can be viewed without third-party requests; all other original-coordinate requests are user initiated and checked against the preserved source SHA-256. Changed provider bytes are rejected. Cross-domain Cα overlays are fully local and preserve canonical UniProt numbering. These derived models are not complete receptors.

Data and code availability

Versioned data downloads, domain APIs, evidence queries and residue correspondence are available without registration. The repository candidate includes the data dictionary, reproduction scripts, source attribution, checksums and correction history. A public repository URL and archival identifier will be supplied after approval; neither has been invented or deposited.

Citation and reuse

Provisional resource title: PWWP Atlas connects structural coverage with domain and ligand evidence, version 1.2 submission-preparation candidate. Authors and an archival citation are pending confirmation. Cite the relevant original data providers and primary assay publications alongside the eventual resource citation.

MIT for original software and CC BY 4.0 for original curation are proposed and await rights-holder approval. This candidate grants no new license. Existing third-party terms continue to apply, including wwPDB CC0, UniProt and official AlphaFold CC BY 4.0, and retained software notices. Public release remains gated by rights and scientific approval.

All source references · US-align · 3Dmol.js · AlphaFold guidance

Maintenance and corrections

The proposed service is anonymous and read-only. Corrections are reviewed, versioned, tested, and released as a new immutable dataset; earlier versions remain archived. A named maintainer, correction contact, institutional owner and commitment to at least two years of public availability must be confirmed before deployment. This local candidate is not a live public resource.