Subfamily resolution of the poly-specific cazy2go families

The poly-specific CAZy families (one family → many GO MF terms) cannot be propagated safely at the
family level. This resolves them to subfamily granularity so each specific activity is attributed
to the right sub-group.

Method (InterPro co-occurrence — dbCAN-sub was inaccessible)

The canonical route is dbCAN-sub's subfamily→EC table, but its server does not serve the data files
from this environment (every candidate URL returns an error page). So subfamilies are resolved from
data we already have: each reviewed UniProt member of a CAZy family carries both its EC(s) and
its InterPro signature(s). Within a poly-specific family F, for each specific GO MF term G
reachable from members' ECs, subfamily_resolve.py finds InterPro
signatures S that co-occur with G at high precision:

precision(S→G | F) = |members of F with S and G| / |members of F with S|   (≥ 0.8)
support            = |members of F with S|                                  (≥ 2)
sig_family_spread  = # distinct CAZy families S appears in   (≤ 2 ⇒ "family-specific")

A family-specific (spread ≤ 2), high-precision S is a sub-family marker for G — a safe,
signature-level attribution of the activity the unsafe family-level term lumped. Each S→G is checked
against interpro2go. Full table: cazy2go.subfamily.tsv.

Results

90 poly-specific (non-CBM) families; 66 resolved to ≥1 signature→GO; 490 signature→GO rows:

Filtering proposals to family-specific signatures (spread ≤ 2, dropping broad superfamily entries
like the GH10 TIM-barrel IPR017853 that are precise within a family but too broad to propagate)
leaves 210 trustworthy proposals across 56 families — signature-level, safe-to-propagate
interpro2go candidates. Examples (all precision ≈1.0, spread=1):

• GH28 IPR050434 → polygalacturonase activity (32 members)
• GT4 IPR056736/IPR056735 → sucrose synthase activity (24)
• GH2 IPR050347 → β-galactosidase, GH3 IPR050288 → β-glucosidase
• GH47 IPR050749 → mannosyl-oligosaccharide 1,2-α-mannosidase (20)
• GH20 IPR029019 → β-N-acetylhexosaminidase, GH38 IPR041147 → α-mannosidase
• GH13 IPR044143 → 1,4-α-glucan branching enzyme (a GH13 subfamily, the canonical poly-specific
  amylase family resolved to its branching-enzyme sub-group)

Interpretation

• The 71 CONFIRMS are the headline validation: an entirely independent signal (InterPro
  co-occurrence within reviewed members) re-derives mappings curators already put in interpro2go.
• The resolution fixes the poly-specific safety problem: where family→GO was unsafe (GH13 → 27
  terms), signature→GO attributes each activity to its sub-group (e.g. branching enzyme vs
  cyclomaltodextrin glucanotransferase via distinct IPRs).
• The 210 family-specific proposals are the actionable, subfamily-resolved interpro2go addition
  candidates.

Caveats

• This is InterPro-signature subfamily resolution, a proxy for the CAZy/dbCAN-sub subfamily
  namespace (which was not fetchable here).
• Precision/support are computed over reviewed members only — small N for niche families.
• spread ≤ 2 is a heuristic; a few legitimate subfamily signatures span 2–3 related families.
• CONFIRMS validate; PROPOSED rows are candidates requiring a human GO/InterPro2GO curator to
  confirm the signature universally implies the activity before any submission.