You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
M3: structure collection for sequence-only and viral input #95
Goal: make M3's structure collection correct for sequence-only (#87) and viral (#88) input.
M3 today (s2f/m3_fold/collect.py, quality.py) takes the proteins M2 selected and collects a PDB structure or an AlphaFold DB model by UniProt accession, accepting AFDB models at mean pLDDT ≥ 70 and masking residues below 70.
What changes
Assumption
Sequence-only input
Viral input
AFDB is keyed by UniProt accession
a novel sequence has no accession: no AFDB lookup at all, so prediction (#54) becomes the main route
many viral proteins are in UniProt, but AFDB coverage of viruses was historically thin; check the viral complex release (24 Sep 2026, #84)
One chain per protein
fine
polyproteins must be cleaved into mature peptides first; a full-length polyprotein model is not meaningful
Monomer models
fine for most bacterial targets
many viral targets are oligomers or complexes (proteases, capsids, spikes); a monomer can hide the binding site
PDB best hit by identity
rarely our own organism
well-studied viruses have many experimental structures; pick by coverage and resolution, not first hit
pLDDT gate 70
fine
viral proteins have more disordered regions; the gate may reject useful partial models (record masked regions, do not drop)
Structure feeds M4 docking
same
a viral pocket may sit at a chain interface — needs the complex
Activity