# PWWP Atlas Discovery and Architecture 1.0.0 ## Scope and versioning This module is a newly derived, read-only index built beside the frozen PWWP Atlas v1.2 scientific dataset. It does not alter the database, earlier analyses, experimental coordinates, assay values, chemical identities, or molecular-viewer transformations. The frozen source snapshot is 24 October 2026. Supplementary UniProt feature responses were retrieved on 26 October 2026; each response has an individual retrieval timestamp, source URL, byte count, entry/sequence versions, and SHA-256 receipt in `sources/uniprot/`. The output is `atlas_discovery.json`. Provenance is recorded in `source_provenance.json`. The web application consumes these files; it does not contact UniProt or perform scientific calculations on a visitor's request. ## Reproduction The scripts require Python 3.11 or later and the standard library. From the application root: ```sh # Optional network acquisition. Successful cached responses are reused. python analysis/discovery_v1/fetch_sources.py # Deterministic local build from the frozen database and saved responses. python analysis/discovery_v1/build_discovery.py # Data and Flask route validation; requires the application requirements. python analysis/discovery_v1/validate_discovery.py ``` The first command retrieves `https://rest.uniprot.org/uniprotkb/ACCESSION.json` for each of the 39 existing Atlas canonical accessions. No additional protein identities are introduced by this module. The second command opens SQLite using `mode=ro&immutable=1`. All outputs stay in `analysis/discovery_v1/`. ## Canonical domain architecture ### PWWP records PWWP domain identifiers, interval endpoints, membership tiers, and source qualifications are copied from the frozen `domains` table. These remain annotation cores, not declarations of complete folded-domain boundaries. Predicted PWWP-like candidates retain their separate membership qualification. Clicking a PWWP segment opens its native Atlas record. The evolutionary analysis uses its own separately justified interval selection; architecture display does not silently replace the frozen annotations with those intervals. ### Additional protein regions Other source annotations are accepted only if the retrieved canonical sequence matches the frozen Atlas sequence character-for-character and the returned primary accession is unchanged. All 39 saved sources passed this exact-sequence gate. The current source's entry and sequence versions remain in its receipt. This permits coordinates to be used on the frozen sequence; it does not claim the annotations are themselves experimentally established. Eligible UniProt feature types are `Domain`, `Zinc finger`, and `Repeat`. A feature must have exact, ordered, 1-based inclusive endpoints within the canonical sequence and must not refer to another isoform. Uncertain features would be recorded in `feature_exclusions` rather than plotted. No features were excluded for endpoint uncertainty in this snapshot. UniProt PWWP descriptions and exact duplicates of frozen PWWP intervals are omitted from the additional track to preserve the native Atlas identity. There are 71 additional eligible feature intervals. Feature-level evidence codes and source indices are retained. The diagram uses a linear amino acid axis. Overlapping annotations occupy separate display lanes without moving their coordinates. Other domain types link to the UniProt source because the Atlas has native records for PWWP instances, not for every domain in the protein. ### Deposited constructs The experimental inventory contains SIFTS-aligned canonical ranges, a deposited polymer sequence, independently observed canonical ranges, sequence differences, and unmapped construct ranges. The displayed bars are the saved mapped canonical segments. They are not inferred complete construct endpoints: purification tags, insertions, unmapped termini, and any unassigned residues remain outside the canonical axis. Observed-coordinate ranges appear as a separate inner band. An example is eligible when its chain has a mapped canonical segment and at least one observed PWWP annotation-core Cα atom in the frozen chain/domain comparison table. One example per native PWWP instance is selected where available, giving priority to frozen experimental representatives, then higher observed core coverage, then smaller mapped length, and finally PDB and chain identifiers. Additional distinct mapped intervals are included up to four examples per protein. This is a reproducible selection of illustrative deposited mappings, not an exhaustive construct catalogue or a domain-completeness determination. Mapped identity, source-recorded sequence differences, and engineering status remain visible in each mapping's detail panel. ### Isoforms The isoform table uses the frozen v1.2 independent PWWP-domain correspondence audit. Only noncanonical models with a passed sequence audit and mapping method `official_UniProt_VSP_edits` are included. There are 103 isoform/domain records. The retained/partial/absent status and source-mapped positions are copied from `alphafold_domain_confidence`; the six correction records keep their official sequence edits and explanatory notes. Deleted cores are not restored by incidental sequence similarity. A displayed minimum-to-maximum span may contain unmapped positions; the mapped residue count and complete position list remain available. Other-domain architectures are not extrapolated to isoforms. ## Literature and event chronology The source is the existing 128-row verified bibliography plus the frozen 230-entry experimental inventory. Each citation retains its individual verification level, which can range from deposition metadata or verified abstract to selected full-text sections. Inclusion in the index does not upgrade that verification level. Records sharing a normalized DOI, or a PubMed identifier when no DOI is present, would be grouped while retaining every source record and verification scope. No records in this snapshot merged under that rule; the literature index therefore contains 128 records. Unknown identifiers are not reconstructed from titles. The original reference-record page remains available. There are 98 dated publication events: 97 PWWP research-context entries and one supporting methodological citation. Publication entries use only the source year. They are not assigned an invented month/day or represented as 1 January. Twenty-nine citations have no publication year and remain searchable in the literature index. One further dated record describes a structural deposition without a journal; it is retained in literature but is not called a publication. The Phenix software-methodology paper (DOI 10.1107/S2059798319011471) appears in the frozen inventory as the secondary citation `REF:PDB:29HG:1`, accompanying 29HG/29HH/29HI. Its recorded title and secondary citation role establish its methodological context. The newly derived display labels it “Methodological citation · not a PWWP finding” and excludes it from the default research timeline. It remains unchanged and searchable in the 128-record literature index, the “All source citations” scope, and a dedicated “Methods context” scope. This is a display classification; neither the frozen citation nor its date is changed. Other cards retain explicit publication types, including reviews, sequence research, computational studies, and deposition-linked bibliography. Every experimental inventory entry supplies two distinct events: | Event type | Frozen source field | Precision | | --- | --- | --- | | PDB deposition | `deposit_date` | Calendar day | | PDB initial release | `release_date` | Calendar day | | Publication | `references.year` | Calendar year | Thus, the event index contains 98 publication-year events, 230 deposition events, and 230 initial-release events. These are recorded metadata events, not 558 distinct discoveries. A PDB entry can be associated with a protein without observed PWWP coordinates; that status is retained. Mixed event views do not infer a within-year chronology for publications whose precise date is absent. Revision dates are not substituted for initial releases. No historical priority claims or functional milestones are inferred. ### Scientific links Protein and PDB links come from the frozen bibliography. Native domain links on publication cards come from curated assay records linked to that reference. Native domain links on deposition cards identify domains with observed core coordinates in that PDB entry. The interface labels these different link scopes. External PDB citations not included in the frozen Atlas open RCSB directly and are marked as external. Ligand/preparation names appear only when a curated assay record associates them with the reference. Compound preparation identifiers, assay IDs, identity status, and evidence class are retained. A modeled CCD component is not substituted for a preparation with unresolved stereochemistry; no affinity is transferred between stereoisomers. The module does not infer activity from structural occupancy or classify every mentioned ligand as a chemical probe. ## Interface and testing scope `/discovery` provides a filterable chronology. `/discovery?view=literature` and `/literature` provide the complete searchable literature index. Filters operate on the entire dataset by query, protein, year, and event type, with explicit ordering and citation scope. The timeline defaults to PWWP research context; the literature index defaults to all source citations. A server-rendered GET form remains usable without JavaScript. The small JavaScript enhancement applies select filters and expands source notes. All visual styling uses the existing Atlas light/dark tokens. The API endpoints `/api/discovery` and `/api/architecture/ACCESSION` are read-only. Discovery downloads use a fixed filename allowlist and return a SHA-256 header. They do not extend the original frozen-data download manifest. Flask route and data validation results are written to `validation.json`; browser interaction and screenshot evidence are reported separately by the application-level QA. ## Scientific limitations - This is a source-indexed research aid, not an exhaustive systematic review or a new claim about the origin of the PWWP family. - Recorded publication, deposition, and release dates cannot establish discovery priority, unpublished experimental dates, or functional milestones. - Canonical sequence identity validates coordinate transfer; it does not certify every UniProt annotation experimentally or settle domain-fold boundaries. - Construct mappings can include source-recorded mutations and exclude unmapped tags or termini; selecting an example does not certify a complete fold. - Isoform evidence covers source-validated PWWP cores only. It does not establish expression, stability, function, or full domain architecture in an isoform. - Missing dates, regions, ligands, or links are missing curation in this snapshot, not proof of biological or bibliographic absence. - Similarity, conservation, shared domain names, and proximity in any tree do not establish identical pockets, histone specificity, chemical selectivity, or therapeutic interchangeability. ## Primary data sources - UniProt REST API: https://rest.uniprot.org/uniprotkb/ (exact per-accession URLs and response hashes are in `source_provenance.json`). UniProt attribution: https://www.uniprot.org/help/license . - RCSB PDB entry pages: https://www.rcsb.org/structure/ (exact entry URLs, source dates, and mapped records are retained in the frozen experimental inventory). - Original papers: exact DOI/PubMed links remain on each literature record. Alvarez-Venegas and Avramova (2012), DOI https://doi.org/10.1186/1471-2148-12-101 , is discoverable in the existing bibliography. Its evolutionary interpretation belongs to the separately versioned sequence analysis rather than this chronology.