# UPF260 sequence-divergent family analysis This directory is the frozen, reproducible analysis of `combined260_remote_plus_sequence_divergent_candidates.faa` performed on 2026-08-21. The public dashboard is staged under `../uniref50_mmseqs2_25339_20260820/site/upf260/` and the relational result store is `upf260_results.sqlite3`. ## Outcome first - 260 input proteins, 256 unique amino-acid sequences. - 237 UPFs at the default MMseqs2 boundary (30% identity, 70% bidirectional coverage, E-value <= 1e-3, set-cover clustering). - 16 families entered profile analysis: 15 by recurrence/context priority and one strict dbCAN rescue. - Final profile classification: one **CAZyme-like** family (`UPF260_0002`), one **other function-associated** family (`UPF260_0036`), and 14 **structurally unresolved** families. - Only one CAZyme-like family is supported by the present evidence. The result was not inflated to the requested 2-5 candidates. `UPF260_0002` is supported by independent evidence from dbCAN CE3, Pfam SGNH/GDSL-related profiles, a PDB70 acetyl esterase match, and a UniRef50 SGNH/GDSL hydrolase hit. Its isolated CE3 region (full-length residues 350-597) has AlphaFold mean pLDDT 92.504 and pTM 0.904. The full-length N-terminal region is predominantly low confidence and was excluded from docking. Template-guided ACT and BOG docking did not recover a pose within 2 Angstrom of the transferred reference pose. Production complex MD is therefore **gated**, not run: a justified native acetylated glycan, experimental activity evidence for the Asp-to-Asn active-site variant, and a reproducible catalytic pose or experimental constraint are needed first. ## Pipeline and fixed decisions 1. **Internal clustering.** Four MMseqs2 sensitivity points were retained: 20%/50% -> 226 families, 25%/60% -> 233, 30%/70% -> 237, and 40%/80% -> 239. The 30%/70% result is the default UPF catalog. 2. **Distribution.** Every family is recorded against M65, M70 and M75 and all observed MAG identifiers. Zero cells are retained in SQLite/TSV so later parameter changes do not alter the denominator. 3. **Observable neighborhood.** Candidate-background proteins on the same sample/MAG/contig and within +/-5 gene identifiers were recorded. This is not a complete genomic neighborhood because original GFF files and full MAG contigs were not available. 4. **Shortlist.** Fifteen families were selected by recurrence, MAG/sample spread, sequence-divergent fraction and observable context. A strict CE3 dbCAN HMM hit rescued `UPF260_0002`, yielding 16 profile families. 5. **Profile evidence.** UniRef50 MMseqs2 hits and HHsearch against dbCAN-fam V9, Pfam-A 35.0 and the PDB70 2022-03-19 snapshot were integrated. 6. **Three-way family class.** Default informative HHsearch evidence requires probability >=95%, E-value <=1e-3, at least 50 aligned columns, and at least 25% query or template coverage. CAZyme profile evidence uses probability >=90% and additionally requires dbCAN template coverage >=35%; strict dbCAN HMM evidence is retained. Significant informative UniRef50 evidence can support the non-CAZyme function-associated class. 7. **Structure and docking.** ColabFold/AlphaFold2 models were summarized at model, region and residue level. The CE3 domain was structurally aligned to 5B5S/5B5L and docked with fixed AutoDock Vina seed 20260821, exhaustiveness 32 and 20 output modes. ## Important interpretation limits - “Structurally unresolved” means that the searched profile databases did not supply a sufficiently strong informative match under the stated rule. It does not mean “no function”. - PDB70 is explicitly the 2022-03-19 HH-suite snapshot; it is not described as current. - The CE3 catalytic assignment is provisional. The template Asp maps to Asn571 in the query, while Ser365 and His574 remain geometrically compatible. This may represent a noncanonical dyad or an inactive variant. - BOG is a crystallographic pocket probe/detergent, not a validated native substrate. ACT is too small to provide an informative affinity ranking. - The phage/host-surface glycan interpretation is a source-grounded hypothesis, supported by BppU-like UniRef50 matches and an observable prophage neighbor, but is not a demonstrated biochemical function. ## Key outputs - `catalog/upf_family_summary.tsv`, `catalog/upf_members.tsv`: 237-family and 260-member catalogs, including separate `query`, `target`, `pident`, `alnlen`, `evalue`, `bits`, `qcov`, and `tcov` columns. - `classification/selected16_three_way_classification.tsv`: frozen three-way profile classification with each evidence metric in its own column. - `hhsearch/*/all_hits.tsv`: complete parsed dbCAN, Pfam and PDB70 HHsearch hits. - `alphafold/results/`: model-, region- and residue-level QA plus rank-1 PDBs. - `docking/results/`: all Vina modes, template-pose RMSD audit and MD gate JSON. - `upf260_results.sqlite3`: normalized, indexed database for downstream tuning. - `database_validation_report.json`: SQLite integrity result, row counts and final SHA-256. Rebuild the database and web payload with: ```bash python3 build_upf_results_database.py \ --site-dir ../uniref50_mmseqs2_25339_20260820/site/upf260 ``` ## Method resources - MMseqs2: - HH-suite: - dbCAN database preparation: - CE3 structural study and 5B5S/5B5L context: - US-align: