Structure-derived candidate profiles¶

Resource bundles v1.1.0 and later contain 33 candidate profiles built from structure-supported protein clusters returned by a capsid search of an EVE catalogue. The set includes profiles with non-capsid functional evidence. The profiles use 5,305 training proteins and belong to the 1,086-profile marker library.

Functional annotations¶

Curated function	Profiles	Training proteins
Putative penton protein	20	4,518
Putative adenain-like protease	4	648
Putative tail tube protein	1	34
Unresolved protease-like protein	1	18
Putative minor capsid protein	2	28
Putative coat protein	2	10
Putative picornavirus-like capsid domain	1	11
Putative capsid protein	1	14
Putative major capsid protein	1	24
Total	33	5,305

Identifiers retain the historical form EVE_CAPSID_NOVEL_NNNN. EVE_CAPSID_NOVEL_0029 is the only profile registered as an MCP, and that assignment remains putative. Its strongest structural reference is a Pleurochrysis polinton-like-virus MCP. Related matches reach NCLDV capsid proteins. These short proteins do not support an exclusive virus-family assignment, so the runtime annotation assigns no marker family by name. A detected protein can still contribute a taxonomic vote through its independent sequence matches. A hit whose references are only EUK-pEVE takes the class of the profile's training EVEs instead (see Profile classes).

The other 32 profiles are not MCPs. Their identifiers and source captions do not confer MCP status. Capsid-type profiles give capsid evidence instead, as described next.

Capsid evidence¶

Twenty-seven profiles count as capsid evidence. They are the 20 penton, two minor-capsid, two coat, one picornavirus-like capsid-domain, one capsid and one major-capsid profiles: 0002 to 0005, 0007 to 0014, 0016, 0017, 0019, 0021 to 0025 and 0027 to 0033. The curated function decides membership; the curation table lists a penton, capsid or coat function for each. CAPSID_EVIDENCE_MODELS in src/virosync/pipeline/phase3/mcp_detection.py holds the list, and a test checks it against the curation table.

A hit to one of these profiles is treated as a viral capsid hit:

Phase 1 lets it seed a candidate region when it validates on a viral or EUK-pEVE reference, and when it has no DIAMOND match and passes the HMM-only gates (validated_novel).
Phase 3 sets has_capsid_evidence on the region when has_mcp is true or any validated or validated_novel hit of the region is such a profile, even if a higher-scoring non-capsid profile on the same protein hides it from the marker count. A hit of another status (supported, unvalidated) gives none, even when it outscores the other profiles. The A1 seed export applies the same rule to the anchors of each seed. The flag supplies the capsid shortcuts of the class-specific acceptance gate.

Graduated hallmark scoring and additive spatial rescue use the exact-model catalogue in virosync/data/hallmark_panels.json. Catalogue structural anchors and capsids are separate roles: a penton or other structural profile does not automatically become an MCP or receive the capsid-alone floor. Only mapped core evidence with a production validation status (validated or validated_novel) can supply graduated marker weight or a spatial witness. Supported and unvalidated hits remain diagnostic records, including hits to catalogue structural anchors. A qualifying spatial witness retains its original validation status and full-gene coordinates. A fully retained witness can pass the native gate with an explicit rescue reason without changing its taxonomy. See Confidence and limitations for the weighting, containment and floor rules.

The catalogue resolves all 33 EVE_CAPSID_NOVEL_* profiles by exact ID. The older profiles VS000652, VS000745, VS000783, VS000788 and VS000791 are exact MCP aliases in Nucleocytoviricota and Polinton-like virus panels; VS000652 and VS000783 also match the Virophages MCP family. Direct MCP-chain and archived PLV-reference matches support these links. Each uses the existing MCP capsid bonus and spatial floors after viral-reference qualification. Aliases count as one MCP family. Runtime MCP recognition and gene-function labels use these exact IDs without assigning an NCLDV lineage. The existing MCP core aliases VS000236, VS000496 and VS000005 also have exact runtime MCP recognition and explicit MCP gene-function labels. Their catalogue roles and panel assignments are unchanged. For fold classification, VS000005 is an exact HK97 alias: a validated hit can infer HK97; an unvalidated hit stays UNKNOWN without independent fold evidence. Generic length or multiple-HMM-domain heuristics do not turn this alias into DJR. The 22 penton profiles share one function: 0019 and 0027 are minor capsid aliases of Penton. All except 0002 belong to the Virophages Penton core family, based on broad direct mavirus minor-capsid matches. 0002 retains an Unresolved_Penton core and supplies support-only Virophages Penton context; its broad adenovirus Penton match does not establish Virophage compatibility at the same level. All 22 supply support-only mCP context in the Polinton-like virus and Maverick panels, matching the existing V18-19 family. Penton evidence can contribute one core but cannot anchor a capsid bonus, capsid-alone floor or capsid spatial witness. Virophages pentons can join three-core witnesses with a MEDIUM floor or partner a capsid MCP within 5 kb for a HIGH floor. A structural match to an adintovirus penton, including 0031, does not assign an NCLDV or PLV type.

The four other capsid or coat profiles have one-family unresolved panels and capsid=false. Each unresolved panel supplies at most one core and contains no scoring context from another family. This keeps uncertain lineage evidence from combining with established panels, but can also withhold credit for real combinations. Protease profiles 0001, 0006, 0015, 0020 and VS000374 supply support-only Protease context in Virophages, Polinton-like viruses and Mavericks. 0026 (assemblin/pUL26N-like protease) and 0018 (gp54 baseplate/tail apparatus) remain excluded from core and context scoring.

VS000375 and VS000151 supply one D5_primase_domain context family in Nucleocytoviricota. VS000146, VS000149, VS000072 and VS000164 supply one mRNA_capping context family there. These partial-domain assignments do not establish complete protein architecture or viral specificity. Support-only links cannot create a core, spatial witness or seed; their families and physical loci are deduplicated with other evidence. Fusion-family exclusions remain in force.

Retained Phase 3 labels also supply exact VOG support: VOG03169, VOG05151, VOG45855 and VOG50158 for Nucleocytoviricota MCP; VOG21003 for A32; VOG04885 for TFIIB; and VOG23436 for PolB in Nucleocytoviricota and Mirusviricota. VOG37123 supplies Virophages Penton and Polinton-like/Maverick mCP context. Partial reference-domain matches support these functional links, not viral specificity. Only the retained label reaches scoring, at its existing profile cutoff and within the final interval; raw unretained hits add nothing.

The putative MCP 0029 maps to the existing MCP family in the Nucleocytoviricota and Polinton-like virus panels with capsid=false. A validated hit contributes core credit, including discounted credit for weak evidence at E <= 0.01. The catalogue flag withholds the graduated capsid term, catalogue capsid-alone floor and capsid spatial witness. Its separate MCP registration in mcp_detection.py still governs MCP annotation, Phase 1 seed confidence, class-vote overrides and output keep rules. It can join ordinary three-core witnesses within 15 kb for a MEDIUM floor. Available controls contain no qualified positive 0029 hit to justify enabling those capsid effects. Catalogue tests establish software behavior, not biological specificity. A viral DIAMOND hit does not add a model to the catalogue.

Explicit support-only links require E <= 1e-5 for scoring context in both Phase 1 and Phase 3. They add coherence only to a panel with a qualifying core; they cannot supply an independent core, catalogue floor or rescue witness. Family and physical-locus matching prevents duplicate context credit.

has_mcp describes the region's marker summary. mcp_gene_ids lists every protein with a validated or validated_novel MCP hit, even when another profile scored higher on that protein. These plotting IDs do not add confidence or alter taxonomy. Both fields refer only to major capsid proteins. The class-vote override, ANI class transfer, jelly-roll classification, Boltz input selection and the Phase 1 high seed confidence still need an MCP. The profile 0029 remains the only EVE_CAPSID_NOVEL_* MCP.

The six other profiles (0001, 0006, 0015, 0020, 0026 and 0018) stay viral markers. They validate and seed like any marker, but they give no capsid evidence.

EUK-pEVE references¶

Resource bundle v1.2.0 relabels the eukaryote-labelled reference hits of the 33 profiles from EUK to EUK-pEVE. These are 3,011 proteins, of which the marker database holds 1,031 and the proteome database 1,892 after dereplication per genome. The key of an affected genome is EUK-pEVE__<assembly>, and its lineage in taxonomy/labels.tsv starts with EUK-pEVE and carries the suffix -pEVE on every named rank. Viral labels on eukaryotic assemblies, such as PPV, keep their class.

ViroSync counts EUK-pEVE as viral and never as cellular or host:

A top-10 hit on an EUK-pEVE reference with at least 25% identity validates a marker and counts as a viral hit in boundary refinement, gene-level taxonomy and region scoring.
When an EUK__ and an EUK-pEVE__ entry share one <genome>|<protein>, the EUK-pEVE row wins the twin rule.
EUK-pEVE hits never enter the host-signature model or the host lineage share.
EUK-pEVE carries no viral family. It sets no NCLDV, MIRUS or PPV flag, joins no dominant-family count and casts no family vote of its own. A marker hit whose references are all EUK-pEVE votes the class of its profile when the profile has one (see Profile classes). Without a class (0002, other markers such as GVOGm0003) the hit stays VIRAL_UNKNOWN, and a region whose viral hits are all EUK-pEVE, in the marker database, the proteome database or both, uses the MIXED rule: it has the likely family MIXED and takes the mixed path of the v2 gate, whatever lineage its marker names or its Phase 1 region class suggest. When a lineage hit exists, the EUK-pEVE hit changes nothing: in the region vote, a gene whose viral hits are all EUK-pEVE casts no vote while any other gene votes for a lineage, apart from the marker vote of a profile class. The A1 seed export applies the same marker votes to the top-10 hits of the seed's validated markers. The taxonomy_best_hits partition counts these genes in the bucket EUK-pEVE; no output class has that name.

Profile classes¶

For 32 of the 33 profiles, one class holds at least 80% of the training EVEs. The counts are the context_PPV, context_NCLDV, context_MIRUS, context_PHAGE, context_CRESS and context_VIRAL_UNKNOWN columns of release-manifests/resources_v1_1_0/capsid_model_curation.tsv, with VIRAL_UNKNOWN in the denominator. The classes come from the 2026-08-08 ViroSync screen of EVEs.

Class	Profiles
`PPV`	29: every profile except `0002`, `0018`, `0026` and `0033`
`MIRUS`	`0026`
`CRESS`	`0033`
`PHAGE`	`0018`
none	`0002` (NCLDV 9 of 15)

STRUCTURE_PROFILE_CLASSES in src/virosync/pipeline/phase3/mcp_detection.py holds the table, and a test derives it from the curation file.

The curation rows that say "EVE context is not a taxonomic assignment" still hold for the profile. The class is used only for a marker hit whose qualified viral references are all EUK-pEVE. A hit with a lineage reference, or with no qualified viral reference, gets nothing from the class. For a hit that qualifies:

The marker votes the class as an ordinary marker (weights 2 and 3), never as an MCP (flat weight 5, override). This holds for 0029 as well.
The class is the marker family in Phase 3 and one marker of that family in the Phase 1 region family. The likely family and the gate use it, so a 3 kb region with only a validated 0008 hit and EUK-pEVE references takes the PPV path of the gate.
The A1 seed export applies the same marker votes, with the same taxonomy lookup, to the validated markers, so A1 and Phase 3 publish the same class for the same evidence. Whenever those rules decide the class, A1 derives the likely family with the Phase 3 function (infer_likely_family); otherwise it keeps the seed family.

Each protein counts once. When several profiles hit one protein, the one that wins the per-protein deduplication of the vote gives the class (an MCP that can vote as one first, then the higher score, then the first hit), in the marker family of Phase 3, in the Phase 1 terminal-repeat children and in A1.

Exception: Phase 1 has no CRESS or PHAGE region family for any marker, so the CRESS class of 0033 and the PHAGE class of 0018 apply from Phase 3 on (the vote, the marker family and the gate). In Phase 1, in its terminal-repeat children and in the region_classification of the A1 export, a hit of either profile gives the region family UNKNOWN. The A1 taxonomy_class and likely family still follow the vote.

Proteome gene records whose hits are only EUK-pEVE cast no class. Hits whose references are all EUK-pEVE vote the profile class when the profile has one. Without a class (0002, other markers such as GVOGm0003) they stay VIRAL_UNKNOWN, and a region whose viral hits are all EUK-pEVE uses the MIXED rule, with the likely family MIXED.

Source groups¶

Source EVE context	Training proteins	Profiles containing that context
Preplasmiviricota (PPV)	5,210	30
PHAGE	31	1
MIRUS	18	1
NCLDV	18	3
CRESS	14	1
Unresolved viral context	14	7

PPV combines polinton-like viruses, virophages, and unresolved PPV calls in the source catalogue. Contexts can overlap within a profile; the last column therefore does not sum to 33. These labels describe the surrounding EVE. They do not establish a profile's taxonomic specificity or functional assignment. ViroSync uses them only through the profile classes above.

Curation and provenance¶

All 33 source profiles are retained. The source caption and curated function are separate fields:

0001, 0006, 0015, and 0020 retain the source caption Product: capsid protein 23K. Their curated function is putative adenain-like protease because the selected reference is an adenovirus protease.
0018 retains the source caption Capsid protein. Its curated function is putative tail tube protein, supported by its strongest BFVD match and chains in the experimental T4 baseplate structure.
0026 retains the source caption Major capsid protein. Its curated function is unresolved protease-like protein. It matches the N-terminal portion of a capsid precursor and has competing experimental protease-fold matches. This evidence does not establish a mature-MCP function or an active protease.

The release curation table records each original identifier, source caption, retention decision, curated function, source-group counts, and profile checksum. The runtime annotation table is models/model_annotations_with_interpro.tsv within the resource bundle.

Pfam annotations use the gathering thresholds from the shipped 937-profile screening library. An empty signature records no qualifying hit among the training members; it is not evidence that a protein lacks domains.

Pre-extracted marker references¶

In resource bundle v1.1.0, pre-extraction yields 3,522 distinct reference proteins. Of these, 230 overlap the existing marker set and 3,292 extend it, giving 3,867,592 marker proteins. The additions comprise 2,997 EUK, 265 PPV, 12 NCLDV, 11 MIRUS, and seven PHAGE references. These are source groups, not model-specificity assignments.

All 33 profiles search the broad reference proteome at a full-sequence E-value cutoff of 1e-5. Each profile searches all 46,765,750 filtered proteins, so its E-values use the full database size. One protein exceeds 100,000 amino acids and is excluded, matching the pipeline's length limit. No additional bit-score cutoff is applied during pre-extraction.

Qualifying hits contribute the full reference protein to marker/marker.faa. Eleven source headers are normalized to existing genome keys for taxonomy lookup; the extraction record preserves their original identifiers. Other headers remain unchanged. A match does not change the reference's source group or establish its function.

Existing marker proteins are preserved. Legacy organism-name suffixes are resolved through the same taxonomy lookup used at runtime, and an existing organism/protein pair is reused only when its sequence matches. Identical proteins from distinct reference organisms remain separate.

Resource bundle v1.2.0 dereplicates the marker database within each genome (see Databases) and holds 2,917,745 proteins. Every pre-extracted reference remains, as itself or through the longest member of its cluster of at least 90% identity in the same genome.

One reference, PHAGE__IMGVR_UViG_8103388301_000001_163, has no detailed lineage entry. It retains its source PHAGE label through the existing runtime prefix fallback. Its detailed lineage remains unknown.

marker/marker.dmnd is rebuilt from the resulting FASTA. The runtime archive contains that database; the source companion contains the matching FASTA. The release extraction record records the input checksums, thresholds, per-model counts, header mappings, and sequence-preservation checks.

Interpretation limits¶

Profile hits must pass the standard ViroSync marker-validation and region filters. An overlapping ordinary protein and its confirmed frameshift rescue can both retain capsid annotations, while their shared marker locus contributes only once to gene counts and confidence. Annotation IDs therefore need not equal the number of independent scoring loci. Structural similarity supports a putative function; it does not establish an active virion protein, a complete MCP, or a viral family.

Recovery of the training proteins tests profile loading and search behavior. It does not measure sensitivity or false-positive rates on independent sequences. The source collection was selected from existing EVE calls and does not provide an independent genome-wide validation set.

The combined library retains new-profile hits for all 5,305 training proteins after Pfam arbitration. Eight proteins from minor-capsid cluster 0027 score best against penton profile 0028; these functional labels overlap in the current profile set. Both profiles give capsid evidence but are not MCPs.

In a tested source locus, 0029 produces a strong profile match, but its viral sequence-reference hits fall below the validation identity cutoff. ViroSync retains that hit as a candidate; it does not grant validated MCP support.