RamaComp

Compare backbone φ/ψ (Ramachandran) plots across two or more protein structures, side by side, in the browser. Drop in PDB or mmCIF files, read the maps together, and export publication-ready figures. Nothing is uploaded — all parsing and geometry runs locally on your machine.

▶ Live: pranavathiyani.github.io/ramacomp — no install, nothing to set up.

Co-developed by Pranavathiyani Gnanasekar & Claude.

---

Why

Most Ramachandran tools (MolProbity, RAMPAGE, RamPlot, the viewers in PyMOL/ChimeraX/VMD) are built around a single structure at a time. When you want to put a predicted model next to its experimental counterpart, or compare two refinements, or look at a handful of homologues together, you end up generating plots one by one and lining them up by hand. RamaComp does that comparison in one view, and exports the whole panel as a single figure.

Features

• Two ways in. Drop/open local .pdb / .mmcif files, or fetch by ID — a PDB code pulls from RCSB, a UniProt/AlphaFold accession pulls the model from the AlphaFold DB.
• Multiple structures at once. Each gets its own plot; choose how many sit side by side (1–5, or auto, and it never crowds them below readable width on small screens).
• PDB and mmCIF. Author numbering preferred, with fallback to label fields. First model only; highest-occupancy altloc kept.
• Reference contours for the six MolProbity residue categories (general, glycine, Ile/Val, pre-Pro, trans-Pro, cis-Pro), each residue scored against its own category. With the Top8000 reference data baked in (see Reference data), the contours and favoured/allowed/outlier calls are the actual MolProbity reference distributions (favoured = 98% contour, allowed = 99.95% contour). Without the data file, RamaComp falls back to a built-in analytical approximation (four categories, indicative percentages only).
• Colour points by residue type, by chain, or by B-factor / pLDDT (a low→high ramp over the B-factor column — temperature factor for crystal structures, confidence for AlphaFold models).
• Interactive. Hover for chain, residue, φ, ψ, class, region, and B-factor/pLDDT. Per-structure favoured/allowed/outlier counts and a clickable outlier list.
• Export a single plot or a combined side-by-side comparison as PNG, JPEG, SVG, or PDF, or the underlying angles as CSV. SVG keeps axes, points, and labels as true vectors (the contour field rides along as a raster layer).

How to use

1. Open pranavathiyani.github.io/ramacomp (or index.html locally).
2. In the Workspace tab, add structures: drag/open files, or type IDs into Fetch by ID (e.g. 1UBQ, 4HHB or P69905, AF-P69905, comma-separated).
3. Set Columns, pick the reference contour set, and choose how points are coloured (residue type / chain / B-factor·pLDDT).
4. Hover points to inspect; open the outlier list under a plot to jump to outliers.
5. Export one plot from the download icon on its card, or tick compare on several (or none = all) and use Export comparison. Pick PNG/JPEG/SVG/PDF for a figure, or CSV for the data.

The About / Help tab documents the method and limitations in full.

Fetching structures

IDs are routed automatically: a 4-character code (1UBQ) is treated as a PDB entry and pulled from files.rcsb.org; a UniProt or AlphaFold accession (P69905, AF-P69905) goes to the AlphaFold DB. AlphaFold filenames are version-stamped (currently …model_v6), so RamaComp asks the AlphaFold prediction API (/api/prediction/{accession}) for the current file URL rather than hard-coding a version. Everything is fetched directly from those public servers to your browser; nothing passes through any RamaComp backend (there isn't one).

How it works

For each residue with an intact peptide bond on both sides:

φ = dihedral( C(i−1), N(i),  Cα(i), C(i)   )
ψ = dihedral( N(i),   Cα(i), C(i),  N(i+1) )

Dihedrals use the Praxeolitic formula (IUPAC sign convention). A residue is only plotted when the C(i−1)–N(i) and C(i)–N(i+1) peptide bonds are both shorter than 2.0 Å, so terminal residues and chain breaks are skipped rather than producing spurious angles.

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

About the contours and percentages

RamaComp scores each residue against a reference distribution, choosing favoured / allowed / outlier by the 98% and 99.95% contours of the reference population — the same convention MolProbity and Phenix use. Two reference datasets can be loaded, selectable from the Reference data dropdown:

• Top8000 (MolProbity) — the grids MolProbity and Phenix ship, so RamaComp agrees with MolProbity when this is selected. This is the default.
• Top2018 — the larger, more recent Richardson Lab dataset, with contours built from ~3M per-residue φ/ψ measurements.

If neither 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.

Reference data

Both datasets are built by one-off scripts rather than committed raw, so provenance is reproducible. Each writes a small JS file loaded by index.html; whichever files are present appear in the dropdown, and the tool degrades to the analytical fallback if none are.

python3 bake_rama.py        # writes rama-top8000.js   (needs internet)
python3 bake_top2018.py     # writes rama-top2018.js   (needs internet + numpy; large download)

bake_rama.py fetches the six Top8000 grids (rama8000-general-noGPIVpreP, -gly-sym, -ileval-nopreP, -prepro-noGP, -transpro, -cispro) from rlabduke/reference_data, computes the contours, quantises each grid to a uint16 base64 blob, and writes window.RAMA_TOP8000.

bake_top2018.py streams the pydangle Top2018 MC per-residue dataset (Prisant 2026; φ/ψ plus the six-class rama_category label per residue), bins φ/ψ per category, smooths on the torus, derives the same contours, and writes window.RAMA_TOP2018. The smoothing bandwidth (SIGMA_DEG, default 3°) is the one tunable; for contours matching the Richardson Lab's published methodology exactly, request their KDE parameters.

Place the generated rama-top8000.js / rama-top2018.js next to index.html and commit them (the older rama-data.js is superseded by rama-top8000.js).

Both Top8000 and Top2018 are licensed CC-BY-4.0 (© Richardson Lab, Duke), confirmed with the lab. References: Williams et al. (2018), Protein Science 27:293–315 (MolProbity/Top8000); Williams et al. (2021), Protein Science 30:170–184 (Top2018); Read et al. (2011), Structure 19:1395–1412 (Ramachandran categories).

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.

Run your own copy

The canonical instance is already live at pranavathiyani.github.io/ramacomp. To host your own (e.g. a fork):

1. Create a public repository named ramacomp.
2. Add index.html to the repository root, along with README.md and (for validation-grade contours) rama-top8000.js and/or rama-top2018.js generated by the bake scripts.
3. In the repo, go to Settings → Pages, set Source to Deploy from a branch, branch main, folder / (root), and save.
4. After a minute it goes live at https://<your-username>.github.io/ramacomp/.

To run it without hosting, just open index.html in any modern browser — no build step or server needed.

A note on dependencies

The page pulls IBM Plex fonts and the jsPDF library from a CDN. PNG, JPEG, and SVG export work entirely offline; only PDF export needs jsPDF to have loaded. If it can't reach the CDN, the rest of the tool still works and PDF export reports the problem instead of failing silently.

Browser support

Tested against current Chromium, Firefox, and WebKit. Uses Canvas and inline SVG; no framework, no bundler — a single self-contained HTML file.

Data sources & credits

Structures are 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). Co-developed by Pranavathiyani Gnanasekar & Claude.

License

Released under the MIT License.