What RamaComp is

A browser tool for reading and comparing backbone Ramachandran (φ/ψ) plots across two or more protein structures at once — experimental structures, predicted models, or both. Everything runs locally: structure files you open are parsed in your browser and never uploaded. Files fetched by ID are pulled directly from the public RCSB and AlphaFold servers to your browser.

Using it

• Add structures by dropping/opening files, or type an ID into Fetch by ID. A 4-character code (1UBQ, 4HHB) is treated as a PDB entry; a UniProt or AlphaFold accession (P69905, AF-P69905) pulls the AlphaFold model.
• Compare side by side; set how many plots sit per row with Columns.
• Inspect by hovering a point for chain, residue, φ, ψ, class, region, and B-factor/pLDDT. The outlier list under each plot is clickable.
• Export a single plot or a combined comparison figure as PNG, JPEG, SVG, or PDF, or the underlying angles as CSV.

How the angles are computed

Dihedrals use the Praxeolitic formula (IUPAC sign convention). A residue is plotted only when both flanking peptide bonds (C–N) are shorter than 2.0 Å, so chain termini and breaks are skipped rather than producing spurious angles. First model only; the highest-occupancy alternate conformation is kept.

Residues are sorted into the six MolProbity categories — cis-Pro, trans-Pro, Gly, pre-Pro, Ile/Val, general (in that precedence) — with cis vs trans proline decided by the peptide ω angle (the CA–C–N–CA dihedral; |ω| < 30° is cis). Each residue is scored against its own category's reference distribution as favoured, allowed, or outlier.

Colouring

• Residue type — general / glycine / proline / pre-proline.
• Chain — a distinct colour per chain.
• B-factor / pLDDT — a low→high ramp over the per-residue value in the B-factor column. For crystal structures that is the temperature factor; for AlphaFold models it is the pLDDT confidence (higher is more confident). The scale is normalised per structure.

About the contours and percentages

Residues are scored against a reference distribution, with favoured = the 98% contour and allowed = the 99.95% contour of the reference population — the convention MolProbity and Phenix use. Use the Reference data dropdown to choose between Top8000 (MolProbity) — the grids MolProbity and Phenix ship, so results match MolProbity — and Top2018, the larger, more recent Richardson Lab dataset. If no reference data file is loaded, RamaComp falls back to a built-in analytical approximation of the basins (four categories, indicative percentages only). For formal deposition or validation use MolProbity or Phenix directly.

Known simplifications

• With the Top8000 data loaded: the full six MolProbity categories including cis/trans-proline (via ω) and Ile/Val. Without it, a four-category analytical fallback (Ile/Val fold into general, cis/trans-proline merged).
• ω is used only to split cis/trans-proline; non-proline cis-peptides and twisted peptides are not separately flagged.
• One altloc per atom (highest occupancy); first model only for multi-model files.

Data sources & credits

Structures fetched from the RCSB PDB and the AlphaFold Protein Structure Database (Jumper et al. 2021, Nature; Varadi et al. 2024, NAR; AlphaFold DB content is CC-BY-4.0). Ramachandran reference distributions are the Top8000 and Top2018 datasets from the Richardson Lab, Duke, used under CC-BY-4.0 (Williams et al. 2018, Protein Science 27:293–315; Williams et al. 2021, Protein Science 30:170–184; Top2018 per-residue data via pydangle, Prisant 2026). RamaComp is co-developed by Pranavathiyani Gnanasekar & Claude, and released under the MIT License. Source: github.com/pranavathiyani/ramacomp.