Drop a PDB or mmCIF file (Boltz output works directly). The structure is parsed, the ligand depicted in clean 2D matched to its bound conformation, and the binding-site residues drawn around it. Everything runs in this tab — the file is never uploaded, because there is nowhere to upload it to.
This is the only outbound request the page ever makes, and only when you ask for it: the entry is downloaded from RCSB (or PDBe) straight into this tab. Files you drop are still never transmitted anywhere.
Bridges are found from the geometry — disulfide, lactam, thioether, lactone or head-to-tail. Declare one by hand if the model does not show it closed.
A ceiling on every distance drawn. Each type also keeps its own limit (H-bond 3.6, salt 4.5, π-stacking 5.6, π-cation 6.0 Å), so raising this past one of those adds no more of that type. π and salt distances are measured between centroids, not nearest atoms, so they read larger than an atom contact for the same residue pair.
0 = a clean textbook depiction. 1 = torsions follow the bound pose as closely as rigid 2D geometry allows.
0 — each label sits against what it touches, no ring. 1 — a clean ring around the ligand or macrocycle. Above 1 the ring opens out.
Off: hydrogen bonds, salt bridges, stacking and the rest are classified and styled separately. On: every residue inside the cutoff gets one dashed line to its closest ligand atom, with that distance. Same residues either way — only the drawing changes.
The archive carries the input file byte-identical, a run log with every build ID, setting and caveat, the interaction table as CSV and JSON, and the figure.
The style file freezes this look — view, conformation,
cutoff, spacing, layout — together with this structure as the reference
to align to. Feed it to ligbatch.py to draw a whole series
the same way. Manual edits are deliberately left out: they name
specific residues.
Click any interaction in the diagram to hide it, or use the list. Additions are measured from the same coordinates and judged by the same criteria the detector uses — the verdict appears below and is written into the run log.
Click a receptor box and a peptide box in the diagram — either order — then pick the atoms. Additions are measured and judged by the same criteria the detector uses. Click any interaction to hide it.
The projection plane the conformation is matched to is a continuous choice, so “how many orientations” depends on how different two layouts must be to count as different — the sweep above reports that honestly. Click a thumbnail to adopt it.